=============================================================================== ===== Following was posted to comp.parallel.pvm ===== ===== and mailed to those who requested PVM from netlib@ornl.gov by email ====== ===== (did not include netlib accesses by ftp or Web or netlib mirror sites)=== =============================================================================== Date: Fri, 6 Jan 1995 19:51:45 -0500 From: pvm-survey@cs.utk.edu Subject: PVM survey Dear Colleague, In order to improve PVM we are trying to understand how people are using the system. Your feedback is important. Could you take a moment to answer three questions. Are you using PVM today? What are your application(s)? How many and what kind of machines have you used with PVM? Any comments and suggestions for improvement are welcome. Please email the information to pvm-survey@cs.utk.edu. Your help is much appreciated. Best wishes, The PVM Project Team =============================================================================== From: bunge@kokopelli.lanl.gov (Peter Bunge) 1. Are you using PVM today? Yes 2. What are your application(s)? Computational Fluid Dynamics, i.e. 3D-spherical high resolution modeling of convection in the earth's mantle 3.How many and what kind of machines have you used with PVM? a) heterogenous cluster (2 SUN,1 HP,1 SGI, 1 DEC-ALPHA) b) homogenous clusters(16 node IBM, 8 node HP, 8 node DEC-ALPHA) c) massively parallel machines (Cray T3D, IBM SP2) --------------------- comments and suggestions: 1) I would like to see a spawning routine, that can take double precision input. This is useful, since I compile my codes with double prec compiler flags and always have problems when spawning my tasks. 2) I would like to see the number of pvm-routines kept limited. ;-) 3) My overall evaluation and experience with PVM ( I have been using it for 2.5 years ) is extremely positive. Setting the defacto standard for message passing and encouraging network computing is one of the most helpful developments in computational physics right now. ---------------------------------------------------------------------- Hans-Peter Bunge Institute of Geophysics and Planetary Physics ( IGPP ) Los Alamos National Laboratory & Geophysics Dept UC Berkeley PHONE : (505) 665 2863 -> office; FAX : (505) 665 3107 EMAIL : bunge@kokopelli.lanl.gov Mail : IGPP , Los Alamos National Laboratory Mail Stop C 305 Los Alamos, NM 87545 ---------------------------------------------------------------------- ------------------------------------------------------------------------------- From: jhh@texaco.com (Joe H. Higginbotham) >Are you using PVM today? I am running PVM as I answer this message. The job is running over the network using 8 RS6K machines and 2 SUN4 machines. The largest number of machines I've run on the network together is 40. These were a mixture of SGI's, SUN's, and IBM's. I generally run from 10 to 20 processors using PVM. Right now the code I'm most interested in running has some problem on the SGI's which is why they are not part of my current virtual machine. > >What are your application(s)? My applications are seismic processing programs. Right now I'm processing data from the North Sea to image a gas field at a depth of 3.5 Km. The image will be used to decide the extent of the field. > >How many and what kind of machines have you used with PVM? I've run 40 machines on a single problem. I use SGI's, SUN4's, and IBM's. > >Any comments and suggestions for improvement are welcome. >Please email the information to pvm-survey@cs.utk.edu. > - On shared memory multiprocessor machines it would be useful to be able to define a memory buffer which all the processors could access. For example if you are modeling waves traveling through a medium such as the earth it is inconvenient to have to keep a copy of the medium paramteres (velocity, density, etc.) in memory for each processor when they all share memory and could read these parameters from the same place. Joe Higginbotham, Texaco ------------------------------------------------------------------------------- From: Dear PVM: This a reply to your survey request. We are a group with offices throughout the Health Center of the University of Pittsburgh including offices in 5 hospitals, the School of Engineering, and the medical school. (1) Yes we are using PVM today and every day. (2) We are running one application at this time. It is conceptually a client/server application. The servers provide access to files which contain neurophysiological data from surgical patients being monitored in the operating room. Data is collected continuously, processed, and then saved to the file about every 40 seconds. Each time new data appears in a file. Any observer process (client) connected to the virtual machine can access the data through the local server process running on the machine on which the data is being acquired. The client processes run constantly, whether they are in use or not. The data acquisition equipment includes a workstation mounted in a wheeled instrumentation cart. When a case is scheduled, a cart is wheeled into the operating room, attached to the Ethernet running through the hospitals, and booted. A script running on the base node (the node where PVM was started) detects the booting node and spawns a pvmd and a server process on it. Using this approach we routinely run the same virtual machine for many weeks at a time between halts. This application using PVM primarily for its message passing and dynamic group capabilities. The dynamic grouping is used to provide several fault tolerance measures. We are currently developing two computational applications which will utilize the same client/server architectural approach. Because our machines are constantly in use but produce mostly spare cycles, we hope to use perhaps 20% - 50% of those cycles with PVM applications. Both applications involve identification and display of time varying data. The simpler one will be used in the operating room to reduce the time required to provide feedback to the surgeon from 30-40 seconds to 5-10 seconds. The second is an image processing application which will be developed on our network and then ported to a network with higher communication bandwidth, perhaps the ALPHA cluster at the Pittsburgh Supercomputing Center. (3) Our virtual machine includes 30 - 40 HP9000/700 RISC workstations and an SGI Personal IRIS. Several of the HP boxes are off site: Mount Sinai Hospital (New York City), George Washington University Hospital (Washington, D.C.), and the offices of Computational Diagnostics Inc., a local private practice firm which provides intraoperative neurophysiological monitoring in several community hospitals in the Pittsburgh area. We also have a network of HP/Apollo MC68040 workstations running the AEGIS operating system which are used in the operating room. We are seeking a PVM port for this operating system. COMMENTS and SUGGESTIONS: We are extremely pleased with PVM and wish to express our gratitude to the developers and the government agencies which support their work. We have had very little trouble and have been able to generate production software for use in our large clinical service is 2 man months based on PVM. We are of course highly concerned with fault tolerance, as our applications are time critical and can be life critical. For the future we would like to have a fault tolerant version of PVM, one which has the which in particular is tolerant to failure of any node in the virtual machine, particularly the base node, i.e. the first one on which PVM is started. Local caching of the virtual machine configuration at every pvmd along with the planned static group function would make the package far more robust in the face of base node failure as well as network segmentation. It would also be helpful, although I suspect very difficult, to provide a mechanism whereby two independent virtual machines could be merged into one without having to halt one of them. Another improvement would be a threaded version of PVM. Finally, for optimization purposes, it might be worthwhile to have a receive-in-place function (there is already send-in-place) rather than having to unpack at the reading end. Thank you again. Don Krieger (412)692-5093 ------------------------------------------------------------------------------- From: Dear PVM: Please add the following to the paragraph in our survey response in which we discuss the life critical nature of our work. : We have been careful to construct the software so that life critical no potentially critical function depends exclusively on PVM, but only of the software n the software running on the local workstation in the operating room. Thanks, Don Krieger ------------------------------------------------------------------------------- From: : Are you currently using PVM? Yes. : What are your application(s)? Mobile robot simulator under X, mobile robots are connect to the world simulator via PVM, also using GNU Scheme with self written extensions to the language to allow PVM to talk to the world interactively from a Scheme top level. : How many and what kind of machines have you used with PVM? Couple of Suns only. : Any comments and suggestions for improvement are welcome. Its wonderful :-) Keep it coming. : Please email the information to pvm-survey@cs.utk.edu. : Your help is much appreciated. Thank you indeed for such well documented software. : Best wishes, : The PVM Project Team cheers, al ------------------------------------------------------------------------------- From: laprade@dw3f.ess.harris.com (Ken Laprade) > > Are you using PVM today? Yes. > What are your application(s)? Course-grained image processing algorithms written in C++. > How many and what kind of machines have you used with PVM? Dozens of Sun's running SunOS 4.1.3 and 5.3, one 8-CPU SparcCenter 2000, a few SGI Indy's and soon a SGI Challenge. -- Ken Laprade Email: klaprade@harris.com Harris Corporation Voice: (407)984-5727 PO Box 98000, MS W2/7744 or: (407)984-6691 Melbourne, FL 32902 ------------------------------------------------------------------------------- From: drahzal@c1south.convex.com (Matthew Drahzal) > > Are you using PVM today? Yes- PVM is portable method to write parrallel apps... > > What are your application(s)? Mostly Financial Analysis > > How many and what kind of machines have you used with PVM? > mainly the Convex Exemplar, but have also used PA-RISC clusters and Convex C-Series... > Any comments and suggestions for improvement are welcome. > Please email the information to pvm-survey@cs.utk.edu. Keep up the good work. ------------------------------------------------------------------------------- From: Dwayne Car > > Are you using PVM today? Yes! > > What are your application(s)? CEM codes. Currently our high frequency codes CADDSCAT and CAVERN both use PVM. They are physical optics and ray trace based codes. > > How many and what kind of machines have you used with PVM? HP only. We may start using it on our Intel hypercube to maintain one version of the code. The most machines I have run on is 62 HP systems at one time. Please do not ask me to locate them. > > Any comments and suggestions for improvement are welcome. > Please email the information to pvm-survey@cs.utk.edu. We would like to use PVM on our low frequency CEM code CARLOS. It was used to win the Gorden Bell award this last year and is used by the EMCC (Electromagnetic Code Consortium). We just can't find an out-of-core LINPACK/LAPACK that works with PVM. Any suggestions? Does one exist? My only frustration during installation... pvm3/lib/pvm claims it will set PVM_ROOT to default to ~pvm3. A minor adjustment makes this work. I also stick the #!/bin/ksh at the top. The tilda does not work in the Bourne shell. My only frustration during use... Many times a host will be okay and the network will go away for just few minutes. It is unfortunate that a few minutes later the host is decleared dead and everything hangs. It would be nice such that even if the computer is declared dead, if it could come back to life and reattach to the virutal machine where it left off. -- Dwayne D. Car Email: m229708@mail.mdc.com Senior Project Engineer Email: m229708@mdcgwy.mdc.com McDonnell Douglas Corporation Email: m229708@aello.mdc.com P.O. Box 516 MC 064 2263 Phone: (314) 232-7103 St. Louis, MO 63166 Fax : (314) 232-7774 ------------------------------------------------------------------------------- From: mhsmith@nas.nasa.gov (Merritt H. Smith) > > Are you using PVM today? > Yes. We have version 3.3.* installed on all of our systems. > > What are your application(s)? > Three-dimensional compressible viscous flow solution over complete aircraft. Numerical aerodynamic optimization of complete aircraft. > > How many and what kind of machines have you used with PVM? > IBM PC (pentium): 4 systems running Free BSD connected by Ethernet. SGI Workstations, HP Workstations: 30+ systems. IBM SP2: 160 nodes. Cray C90: Used PVM in place of multi-tasking (works well, but uses too much memory to be useful) ------------------------------------------------------------------------ |Merritt H. Smith | Bug-free code, | |Applied Computational Aerodynamics Branch | Rapid convergence, | |MS 258-1 | Peace on Earth, | |NASA Ames Research Center | A dependable Jaguar. | |Moffett Field, CA 94035 | | | | | |mhsmith@nas.nasa.gov (415)604-4493 | Is that too much to ask? | |http://www.nas.nasa.gov/~mhsmith/home.html | | ------------------------------------------------------------------------ ------------------------------------------------------------------------------- From: ptn@chevron.com (Phuong T. Nguyen) >Are you using PVM today? Yes. > What are your application(s)? With MCNP code from Los Alamos Lab. > How many and what kind of machines have you used with PVM? IBM RS/6000 ~7 HP 735 2 ------------------------------------------------------------------------------- From: dianne@cbi5.cbi.com (Dianne Marsh) > Are you using PVM today? Yes. > What are your application(s)? Human Genome Project -- Sequence Analysis > How many and what kind of machines have you used with PVM? We use SUN workstations but are planning to port to other UNIX environments. It is also likely that we will incorporate some other architectures (e.g. MasPar). We have combined Solaris 1 and Solaris 2 machines and have used up to 20 machines in a distributed fashion. ------------------------------------------------------------------------------- From: Are you using PVM today? No, but I've used it recently. What are your application(s)? Optimization (expected values) and PDE's. How many and what kind of machines have you used with PVM? 4 to 6 Sun workstations. Any comments and suggestions for improvement are welcome. Installation was a bit clumsy for a user who's not a systems person. The manual didn't have a complete (or any, in some cases) description of certain commands. The daemons on host machines were difficult to exorcise. {It's still a great addition to the modeling community. Keep up the good work.} ----------------------------------------------------------------------------- James E. Giles Phone (615) 632-1980 Fax (615) 632-1840 TVA Engineering Lab, LAB 1A-N x.400:/c=us/admd=attmail/prmd=gov+tva/ P.O. Drawer E s=Giles/g=Jim/dd.id=idvi8.office/ 126 Pine Road Internet:tva!idvi8.office@mhs.attmail.com Norris, TN 37828 ----------------------------------------------------------------------------- ------------------------------------------------------------------------------- From: meiji!kendo!adc@uunet.uu.net (Alan Cabrera) > > Are you using PVM today? Yes. > What are your application(s)? Parallel path integration. > How many and what kind of machines have you used with PVM? 50 SPARCstations 2, 10, and 20s. Alan D. Cabrera Email: adc@sanwaBGK.com Sanwa Financial Products Co., L.P. Phone: (212) 407-3591 55 East 52 Street - 26th Floor FAX: (212) 758-9552 New York, NY 10055 ------------------------------------------------------------------------------- From: PVM Survey Reply: > Are you using PVM today? Yes. > > What are your application(s)? We are using it to parallelise Computed Tomography (CT) Reconstruction based on algebraic reconstruction technique (ART). > > How many and what kind of machines have you used with PVM? 3 Sun SparcII running SunOS 4.1.3 3 Sun SparcII running Solaris 2.3 2 HP-UX. Thank you, Binu John. ------------------------------------------------------------------------------- From: sanz@lace01.lerc.nasa.gov (Jose Sanz) Please, disregard previous e-mail if you received it. Yes, I use PVM now on a routine manner. My application is " Aerodynamic Inverse Design and Analysis for a Full Engine". I use PVM to run this application on two CRAY C90s. My host is an IBM RS6K-590. I use PVMe on an IBM SP2-160 nodes at NASA Ames. After a problematic installation of PVMe on the SP2 ( it was creating a lot of orphan jobs) now runs excellently. The use on the C90s is rather interactive, whereby I spwan processes from my RS6K host on either C90 as needed. Mostly, this processes run almost independently of each other and with practicly no limit on the number of processes that I am running. ( Among other things PVM has resulted to be an excellent scheduler). On the SP2 I run parallel jobs that require large internode comunications. SP2 runs those jobs on dedicated time. The internode comunication overhead I fund to be very reasonable for the problem I am solving. Congratulations for an excellent piece of work. I can expand, or send you some results on the SP2 if you are interested. J. Sanz NASA Lewis Research Center (216) 433-5917 ------------------------------------------------------------------------------- From: ytassa@fluids4.rdd.lmsc.lockheed.com (Y. Tassa) I am using the PVM and the applications are flow field,and fluid/structure codes primerily two kinds of machines Dec 5000 workstations and SGI . I would like to get detailed doc. on the recent PVM version. Y.Tassa ------------------------------------------------------------------------------- From: Terry Purnell 1) Are you using PVM today? Yes. 2) What are your application(s)? Real-time distributed interactive combat simulations. 3) How many and what kind of machines have you used with PVM? 20+ Sun (SPARC 10, LX), SGI (Indigo, Elan, Onyx), SGI Challenge Cluster, CM-5, SP2. 4) Any comments and suggestions for improvement are welcome. I frequently have to run demonstrations on experimental network and hardware configurations. My team has been actively developing distributed processing applications using PVM for over a year and the most difficulty we have had associated with PVM is getting it to start-up properly on the distributed architecture under the experimental network. Since I am required to develop software for use by researchers who aren't very familiar (or patient) with distributed processing what we could really use are good tools for diagnosing PVM start-up problems. Please email the information to pvm-survey@cs.utk.edu. Terry Purnell U.S. Army Research Laboratory ------------------------------------------------------------------------------- From: baker@jake.plk.af.mil (Lou Baker) Are you using PVM today? YES. What are your application(s)? PSPH, a parallel smoothed particle hydrodynamics code. How many and what kind of machines have you used with PVM? Mostly on SP-2 at Maui hpcc. Have run problems with approx 1/2 million particles for many hours simulating large impacts and explsions. Any comments and suggestions for improvement are welcome. Daemons die when NFS at maui goes down. This creates problems as the NFS ons the SP-2 seems to be somewhat unstable. The result is the program dies (with Signal 15). However, the job appears to be suspended, with resources tied up. It appears(?) that pvmcleanup may not cancel/remove all the suspended processes, thus inconveniencing other users. (The last two sentences are based upon the report of another user.) ------------------------------------------------------------------------------- From: mwtraha@sandia.gov (1241 Michael W. Trahan) Are you using PVM today? Yes. What are your application(s)? Particle-in-a-cell codes (physics). How many and what kind of machines have you used with PVM? IBM RS-6000 2 to 4 Sun SPARCstation 10 and 20 10 to 15 Sun SPARCserver 1000 8 CPUs IBM SP2 2 to 256 nodes Intel Paragon 2 to 128 nodes Any comments and suggestions for improvement are welcome. The ability to use IBM PC-compatibles in the virtual machine would be useful, especially during development. ------------------------------------------------------------------------------- From: Burke Murray Are you using PVM today? Yes What are your application(s)? Computational Fluid Dynamics (Panel Method) How many and what kind of machines have you used with PVM? Only the Cray (YMP to T3D) so far, but I hope to use PVM with networked workstations in the future. ------------------------------------------------------------------------------- From: krishnan@beavis.bnpi.com (Krishnan Ethirajan) Are you using PVM today? Yes. What are your application(s)? Quantum Mechanics/Statistical Mechanics - Molecular Modeling. How many and what kind of machines have you used with PVM? 30 Machines, IBM RS/6000. ------------------------------------------------------------------------------- From: jmckir@uwaf.nswc.navy.mil (John B. McKirgan) Are you using PVM today? Yes. What are your application(s)? I am using an application developed by the Department of Energy (Sandia National Labs). The application is called PCTH. It is a eulerian hydrocode we use to model shock physics and impact dynamics problems. How many and what kind of machines have you used with PVM? The only kind of machine has been Hewlett Packard 700 series workstations. The number of machines varies from 4 to 8. Any comments and suggestions for improvement are welcome. Let me start with my compliments. PVM so far as been a solid package that installed with ease. The only problem I had was in compiling the example problem "xep". I did not have some of the include files (Toggle.h, Form.h etc.) Though, in all honesty, I did not spend a great deal of time on it. My main concern with PVM is it's continued funding. I am now at a point in planning future hardware purchases where I have to decide whether to pursue a low-end massively parallel machine "OR" continue to accumulate workstations. My budget does not allow for both. Much of the software I have been seeing that does my kind of problems, is developed on MP machines like the Intel Paragon. If the software isn't ported to PVM, I've got a problem. Further, I'm not a big enough player to force these software developers to make the ports. I have spoken to my hardware vendors (HP, Digital). My sales reps are not familiar with PVM; which is disturbing. I guess you could summarize this as a technology transfer concern. For now, PVM is the way for me to run problems to big to fit on one machine. ------------------------------------------------------------------------------- From: ram@eniac.cf.com (Ram Pandit) Subject: Re: PVM survey Following are the answeres to your questions : 1. Currently, we are not using PVM. I used it to implement a Branch-and Bound algorithm for learning purposes. However, we are excited about PVM and whenever the opportunity arises, we shall use it. 2. Most of applications are large-scale mixed integer programming problems. For example, Airlines Fleet-Planning, Truck dispatching etc. 3. At present we have two RISC/6000-580 machines hooked together. So far, I have a very limited experience with PVM to really offer you any significant suggestion. However, thanks for PVM its a great tool. ----------------------------------------------------- Ram Pandit Sr. Systems Engineer Intelligent Systems Engineering Consolidated Freightways, Inc. P.O. Box 6696 Portland, OR 97208-6696 Wk: (503) 499-3114 Fax: (503) 499-3947 ------------------------------------------------------------------------------- From: pbd@winternet.com (Paul Dokas) > Are you using PVM today? yes > What are your application(s)? learning to program in a parallel environment > How many and what kind of machines have you used with PVM? 3 machines: Sun, NetBSD on a PC, SGI paul ------------------------------------------------------------------------------- From: "Maciej Kormicki - EECS (EE476)" > > Are you using PVM today? > Yes, I am using PVM. > What are your application(s)? > I am using PVM for my research which is "parallel logic simulation on a network of workstations using PVM" > How many and what kind of machines have you used with PVM? > So far I used simultaneously 4 machines. I am using HP workstations. ------------------------------------------------------------------------------- From: og845@carbon.pnl.gov (Edoardo Apra) I am not using PVM at the moment, but I could use it again in the future. I used pvm3.2 with Crystal92 (a solid-state quantum chemistry code) I used pvm with the following architectures: RS6K, HP, SGI, DEC alpha, Convex, Cray C90 Regards, Edoardo Apra` ------------------------------------------------------------------------------- From: Jyan Hong-Wei > Are you using PVM today? No. > > What are your application(s)? Try to build up a distribute and transparent computation environment for our users, who are not familiar with computer. > > How many and what kind of machines have you used with PVM? Currently, we only got 9 workstations and mainframe to run PVM. Those include 3 HP, 1 Sun, 1 SGI, 1 Dec, 1 Dec-Alpha, and 2 Cray. After we can understand the PVM totally, we will install it over our whole network system. Thanks. -- +----------------------------------------+-------------------------------+ | Name: Hong-Wei ( Howard ) Jyan | | | e-mail: howard@gwya.cwb.gov.tw | "A city is a large community | | Org.: Central Weather Bureau / CWB, | where people are lonesome | | Taiwan, ROC. | together..." | | Div.: System Control Section. | | | Phone: +886-2-3491288 | | +----------------------------------------+-------------------------------+ ------------------------------------------------------------------------------- From: Larry Wickstrom Are you using PVM today? No. What are your application(s)? Video Compression How many and what kind of machines have you used with PVM? 5 SGI Indigo machines ------------------------------------------------------------------------------- From: Steve Johnson Are you using PVM today? Yes What are your application(s)? Monte Carlo simulations of 2D and 3D magnetic systems How many and what kind of machines have you used with PVM? 32 Sun SS10 32 HP 735 Steve Johnson E-mail: Steve.Johnson@math.tamu.edu Unix Systems Manager Phone: (409) 845-0261 Dept of Mathematics FAX: (409) 845-6028 Texas A&M University College Station, TX 77843-3368 ------------------------------------------------------------------------------- From: hicup@fatman.tamu.edu (Hagop Barsamian) To whom it may concern, In response to the three questions: -- I am currently not using PVM. -- I have used PVM on an HP Apollo series (720) and a SUN Spark 10 workstations. From: Stas Polonsky >Are you using PVM today? >What are your application(s)? >How many and what kind of machines have you used with PVM? >Any comments and suggestions for improvement are welcome. >Please email the information to pvm-survey@cs.utk.edu. >Your help is much appreciated. Q1: Yes. Q2: Optimization of superconductor digital electronics circuits. Q3: 3 HP 700, 1 SGI Indy -- +-----------------------------+-----------------------------------------+ | Stas V. Polonsky, Ph.D. | Internet: stas@rsfq1.physics.sunysb.edu | | Physics Department, | Voice: (516)632-8060 | | SUNY - Stony Brook, | Fax: (516)632-8774 | | Stony Brook, NY 11794-3800 | | | U.S.A. | | +-----------------------------+-----------------------------------------+ ------------------------------------------------------------------------------- From: rnagaraj@cne.gmu.edu (Sai Ravi Nagarajan) Are you using PVM today? Yes. What are your application(s)? Distributed scheduling algorithms, performance measurements, etc. How many and what kind of machines have you used with PVM? Upto a maximum of 32 workstations. Usually use Sun and HP workstations. Ravi Nagarajan. Dept of CS, George Mason University, Fairfax, VA 22030. ------------------------------------------------------------------------------- From: Mohamed Iskandarani > > Are you using PVM today? Yes I am. I do not use it on a daily basis, but we have a code that relies on PVM. > What are your application(s)? We have a spectral element code to solve the shallow water equations. Our main focus is basin-scale ocean modeling. > How many and what kind of machines have you used with PVM? The code runs on the nCUBE 2 and on the T3D. The orginal parallelization was carried out on an nCUBE 2 since our university owns one, and we have unlimited access to it at no charge. The message passing was based on the nCUBE's native message passing library. We ported the code to the T3D and decided to base the message passing on PVM. Our code is not 100% portable since we rely on Cray call (fast_send and fast_receive) to bypass the slow PVM calls. ------------------------------------------------------------------------------- From: Hemant Rotithor 1. yes 2. distributed branch and bund 3. cluster of alpha stationsdecstations ------------------------------------------------------------------------------- From: "William T. Rankin" > Dear Colleague, I guess you mean me... :-) > In order to improve PVM we are trying to understand how > people are using the system. Your feedback is important. > Could you take a moment to answer three questions. No problem. > Are you using PVM today? Yes. In fact, as I type this I am. > What are your application(s)? Personally, I have used PVM for the following applications. 1) Molecular Dynamics - As part of an NSF "Grand Challenge" Grant, we have used PVM to implement various multipole algorithms for solving large N-body interaction problems. The resulting libraries have been succesfully integrated into the NAMD Moleculer Dynamics package from UIUC (which also runs under PVM) Hopefully, in the new few months, we will have the XPLOR MD package from Yale use these codes. 2) As part of my Masters research, I implemented distrributed codes to identify numerical sequences known as Golomb Rulers. This search could be likened to a prime number search. I implemented a fault tolerant, restartable parallel code which ran for over three months on a network of 27 workstations and discoversd a new unknown sequence. 3) A proof-of-concept project implementing Active Messages under PVM using Unix signals. In addition to my own work, I support the installation ov the PVM package for the EE department here at duke, where it is used by gradutes and undergraduates for classes and research. > How many and what kind of machines have you used with PVM? I have run PVM on the following platforms, and use most of them regularly. 1) A cluster of Sun workstations under Solaris 2.3 and ethernet 2) A group of IBM RS6000 3) A cluster of HP 735's with ATM 4) A Cray T3D (32 proc and a 512 proc) 5) A 486 clone under Linux. (My home machine) > Any comments and suggestions for improvement are welcome. (Active Messages ;-) > Best wishes, > The PVM Project Team Regards, -bill ---- / __/ / / bill rankin / / / wrankin@ee.duke.edu ___ / / / / philosopher/coffee-drinker / / / / / molecules-R-us / / / / / _______/ __/ __/ __/ "All I wanna do is have a little fun before I die." - Sheryl Crow ------------------------------------------------------------------------------- From: BHASKAR We are using PVM in our Distributed systems course for developing simple projects. We use LINUX on PCs, RS/6000, and possibly VMS and Ultrix on VAXs. Will send suggestions as I - professor of CS in Winona State University, CS Dept. - put them together. PVM has been quite useful for us - thank you. Sudharsan Iyengar (507) 457-5539 ------------------------------------------------------------------------------- From: sliu@spectra.eng.hawaii.edu (Sheng Liu) > Are you using PVM today? No. But I used PVM last spring. > > What are your application(s)? Vector quantization using Kohonen self-organizing feature map. > > How many and what kind of machines have you used with PVM? about 10 Sun workstations > ------------------------------------------------------------------------------- From: "Dr. Eric C. Frey" > > Are you using PVM today? Yes. > > What are your application(s)? We are doing 3D image reconstruction for single-photon emission computed tomography. > > How many and what kind of machines have you used with PVM? > We are using DEC Alphas and DECstations. We have a few stardent computers but haven't been able to get it to run on them. We recently bought an HP computer, but haven't ported our stuff to it yet. > Any comments and suggestions for improvement are welcome. > Please email the information to pvm-survey@cs.utk.edu. > My student has been doing the programing so I can't make a lot of comments about this. I will forward it to him. =============================================================================== Eric C. Frey Phone: 919-966-6729 Research Assistant Professor Fax: 919-966-2963 Department of Biomedical Engineering and e-mail: frey@bme.unc.edu Department of Radiology The University of North Carolina 152 MacNider Hall, CB 7575 Chapel Hill, NC 27599 ------------------------------------------------------------------------------- From: > > Are you using PVM today? Yes. > > What are your application(s)? Robotics and machine vision, having a distributed pvm program govern many real time processes controlling robots and vision HW. > > How many and what kind of machines have you used with PVM? Sun sparc 2000 (8 proc), 12 & 8 proc SGI 480, and single processor servers and workstations in our network. > > Any comments and suggestions for improvement are welcome. The shared memory version for Sun solaris used shmemget with ID #1 or 2 or something stupid like that. Of course we had other programs just as badly written using the same ID#, inadvertedly adressing the same memory. Caused spectacular crashes before we corrected it. Better debugging aids are always welcome. We haven't been able to spawn pvm processes in a debugger window for some unknown reason. An email address list of users who struggle with real time applications, and maybe some sample real time type software for different systems availible by ftp would be great. /m ------------------------------------------------------------------------------- From: jbm@speedy.login.qc.ca (John McCluskey) > > Are you using PVM today? Not at the moment. In the future, perhaps. > > What are your application(s)? I took a class at McGill University in Parallel Processing, and used PVM for a programming assignment where I implemented the Chandy-Misra-Hauss deadlock detection algorithm. It was surprisingly easy with PVM. When I get more time, I'll probably use PVM for implementation of parallel Monte-Carlo simulations on the workstation network where I work. > > How many and what kind of machines have you used with PVM? single host Intel 486 pc-clone running Linux. ------------------------------------------------------------------------------- From: "Steven M. Gallo" > > Are you using PVM today? > Yes, but currently only a specialized version that Cray Research has implemented for use on their MPP T3D. > What are your application(s)? Molecular structure determination. > > How many and what kind of machines have you used with PVM? > Cray T3D, but I have been considering downloading it to test on a network of Linux boxes. Steve Gallo ------------------------------------------------------------------------------- From: attilio@di.unito.it (Attilio Giordana) Are you using PVM today? YES What are your application(s)? We are developing distributed genetic Algorithms. In particular we have a learning system "REGAL" which is based on PVM. We are distributing it by ftp "pianeta.di.unito.it (anonymous) How many and what kind of machines have you used with PVM? We use our system in two sites: (a) a LAN of several SparcStations. (b) a CM5 in Paris (through internet) Any comments and suggestions for improvement are welcome. We found the system quite delicate under CM5. If you are interested we could start a discussion. In any case REGAL at least works on it!!!! ------------------------------------------------------------------------------- From: yuexia@spectra.eng.hawaii.edu (Yuexia Liao) > > Are you using PVM today? No. I used it last year for my master thesis > > What are your application(s)? compare load balancing schemes > > How many and what kind of machines have you used with PVM? 20, including SUN Spac, HP300, HPPA workstations ------------------------------------------------------------------------------- From: David Grimshaw Dear PVM Project Team, Thank you for soliciting my opinion. Unfortunately, I don't have much information to send since I am a rank newbie for PVM. We have so far not taught any parallel computing at Ryerson Polytechnic University (Toronto). I obtained PVM after seeing the news group on netnews. Started 'fooling around' with it Dec 94. To answer your questions, 1) use: teaching (potentially) 2) applications: student undergrad thesis projects (potentially) 3) Installations i) Sun Sparcstation multiprocessor (4) system SunOS 5.3 (Seems to only like running one pvmd3 per processor) ii) Sun single processor Sparc with SunOS 5.3 iii) IBM RS6000 Aix 3.2, one processor. The above systems are at the University and linked by a fibre optic backbone. Have run some of the PVM demos. iv) i486 running Linux v) i386 running Linux These two are linked (at home) via an ethernet connection. Regards, David Grimshaw ------------------------------------------------------------------------------- From: Anshu Aggarwal > Are you using PVM today? -- I was till last month > What are your application(s)? -- 1) Benchmarking applications to measure communication timings between processes. 2) Large scale aerodynamic code to study the effects of fluid turbulence on aircraft structures. > How many and what kind of machines have you used with PVM? -- SPARC Stations, Dec ALPHAs, KSR-1, Cray T3D, CM-5, Intel Paragon. > Any comments and suggestions for improvement are welcome. > Please email the information to pvm-survey@cs.utk.edu. -- There are numerous issues that arose upon porting of applications to the multiprocessors, primarily dealing with how processes were spawned on the different processors and with the pecularities of the implementation of PVM on that machine (PVM was not completely portable from one multiprocessor to another). We also did not have much success in distributing the applications over the different multiprocessors and having them communicate with each other. It was also unclear how good the implementation of PVM was on the different multiprocessors since sometimes there was a significant difference in performance between a PVM program and a program written using the native message-passing calls. If you have any further questions, please dont hesitate to contact me at anshu@cs.colorado.edu -anshu ------------------------------------------------------------------------------- From: Al Bessey > > Are you using PVM today? Yes, in relation to my masters thesis. > What are your application(s)? Parallelizing the Lagrangian Particle Dispersion Model. Basically, the code tracks a cloud of pollutants through the atmosphere, given atmospheric data from RAMS. (Regional Atmospheric Modelling System.) > How many and what kind of machines have you used with PVM? I have used an isloated ethernet network of 7 IBM RS/6000 workstations, and am currently trying to port the code to the Intel Paragon. The port is being held up by my code, not PVM. > Any comments and suggestions for improvement are welcome. > Please email the information to pvm-survey@cs.utk.edu. There are many codes still using PVM 2.4 and you should continue to publish the 2.4 to 3.x.y conversion sheets, possibly as a section in the users' guide. I was lucky enough to have an old copy of the conversion sheet in an old folder, but I would bet many are not that lucky. BTW - I received 4 copies of this message. Al Bessey -- +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ For lack of a Dream...his life was lost....... When a man knows not the harbor for which he is headed, no wind is the right one for his sails. +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ | Al Bessey - Intel PSE | E-Mail: bessey@ssd.intel.com | | California Institute of Technology | Office: (818) 356 - 0335 | | 1201 E. California Blvd | FAX : (818) 584 - 5917 | | M/S 158-79 | Pager : (800) SKYPAGE Pin:870-3437 | | Pasadena, CA 91125 | | +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ ------------------------------------------------------------------------------- From: alves@mercurio.uc.pt (Alexandre Alves) > >Are you using PVM today? I am not a common PVM user: I am developing a PVM API for MS Windows (16/32). >What are your application(s)? We will see in the future, but now I am thinking just about benchmarks. >How many and what kind of machines have you used with PVM? Just a few PCs (80x86 based). Best wishes, Alex ------------------------------------------------------------------------ Alexandre A. R. Alves Departamento de Eng. Informatica tel: 351.39.7000000 (global) Universidade de Coimbra tel: 351.39.7000072 (direct) Polo II tel: 1236 (local) Vila Franca- Pinhal de Marrocos fax: 351.39.701266 3030 COIMBRA - PORTUGAL email:alves@pandora.uc.pt ------------------------------------------------------------------------ ------------------------------------------------------------------------------- From: j killough We are using PVM for parallel finite difference codes for flow in porous media. The primary platforms are IBM/RS6000's of various sorts (550,560, 590). The only complaint I have is that at least in version 3.1, it is not straightforward to redirect the output from the parallel tasks. I would prefer the ability to direct the output from a parallel task to any given dataset. This is particularly useful in debugging. Is the x version of pvm stable at this point? When I discussed this with someone in July, there were still a few bugs. Thanks, John Killough ------------------------------------------------------------------------------- From: Ahmed Ouenes Response to pvm-survey 1)Yes I am using PVM today more than ever 2)Geoscience applications Reservoir Characterization Reservoir History Matching using Numerical Simulators Parallel Optimization Algorithms 3) I used 6 IBM RISC 600 (590), HP 9000-735, SUN SPARC 10 I have comments but they are too long to write in an e-mail I have also couple of papers describing briefly my applications and the way I am using PVM. Feel free to use these references for anything that may support and encourage the use of PVM. Ref 1. Ouenes Ahmed, Srinivasa Bhagavan, Peter H. Bunge, and Brian Travis: Application of Simulated Annealing and Other Global Optimization Methods to Reservoir Description: Myths and Realities paper SPE 28415 presented at the Society of Petroleum Engineers Annual Technical Conference and Exhbition, New Orleans, 25-28 Spetember 1994. Ref 2. Ouenes Ahmed, William Weiss, Ahmad J. Sultan, and Jabed Anwar: Parallel Reservoir Automatic History Matching Using a Network of Workstations and PVM paper SPE 29107 presented at the 1995 Society of Petroleum Engineers Symposium on Reservoir Simulation, San Antonio TX February 12-15. Good Luck PVM team, I am looking forward to see more good stuff Dr. Ahmed Ouenes | Petroleum Recovery Research Center | New Mexico Tech | Socoro New Mexico NM 87801 | ------------------------------------------------------------------------------- From: Tuneyasu Komiya > Are you using PVM today? Yes. > What are your application(s)? We are using PVM to implement distributed parallel Scheme and Common Lisp. > How many and what kind of machines have you used with PVM? Up to 45 SPARCstations and a LUNA88K2. -- Tsuneyasu Komiya Toyohashi University of Technology E-Mail: komiya@katura.tutics.tut.ac.jp ------------------------------------------------------------------------------- From: Milan Hodoscek > > Are you using PVM today? Yes > > What are your application(s)? Molecular dynamics CHARMM (see http://www.ki.si/parallel.html) > > How many and what kind of machines have you used with PVM? HPPA to develop program for CSPP. CSPP is hard to deal with! > > Any comments and suggestions for improvement are welcome. Great library. But I miss some more simplicity: The way CHARMM was parallelized is by using four routines: Get number of nodes Get number of my node send receive So it usually takes few hours to get it run on any parallel platform, but it took me few days to read the whole documentation (which is very well written - I read PVM book - thanx) and copy examples from the book to the program. Sincerely yours -- Milan Hodoscek (http://www.ki.si) ------------------------------------------------------------------------------- From: "Jeff A. Earickson" > > Are you using PVM today? Only occasionally, as I have to maintain it. > > What are your application(s)? Usually just the pvm examples. > > How many and what kind of machines have you used with PVM? Sun and Paragon. ------------------------------------------------------------------------------- From: moyers@mcdgs01.cr.usgs.gov (Jamie Moyers) > Are you using PVM today? Yes. > What are your application(s)? The USGS is scanning its 1:24000 scale topographic maps at 500 dpi. These maps are then downsampled and georeferenced to a UTM grid. We're using applications built on PVM to perform the georeferencing and browse image generation. > How many and what kind of machines have you used with PVM? Almost exclusively Data General Aviions. We have over 100 workstations at this site. These are split up into clusters of 10 workstations for the PVM applications. > Any comments and suggestions for improvement are welcome. PVM is great! We've been able to develop some very useful systems based on it. Keep doing what you're doing. ------------------------------------------------------------------------------- From: Limor Schweitzer I'm using, at times dmake which is a parralel make based on pvm. I've given a 3 day course on pvm. Customers of ours are developping air-flow simulation on Dec-APX clusters, using pvm. Hope this helps, and I would appreciate seeing some of the results of your survey. Limor Schweitzer (limor@xpert.com) Xpert Unix Systems LTD. ------------------------------------------------------------------------------- From: vb1890@PLAY.CS.NYU.EDU (Victor Boyko) pvm-survey> Are you using PVM today? Yes. pvm-survey> What are your application(s)? Computer simulation of atomic structure of the high-temperature superconductor Y Ba Cu O. pvm-survey> How many and what kind of machines have you used with pvm-survey> PVM? Usually about five SGI5's. Sometimes SUN4's and RS6K's. -Victor ------------------------------------------------------------------------------- From: "Fedor G. Pikus" > > Are you using PVM today? Yes. > > What are your application(s)? Monte-Carlo modeling of stronly interacting many-electron systems (quantum dots, coulomb liquids). > > How many and what kind of machines have you used with PVM? Cluster of RS6000 with Ethernet and FDDI, Meiko, Paragon. > > Any comments and suggestions for improvement are welcome. Write a working version for Paragon!!! And global operations wpould be nice. -- Fedor G. Pikus 6113 Broida Hall 312 Ellwood Beach Dr. #18 Department of Physics Goleta, California 93117 University of California home phone: (805) 968-3361 Santa Barbara, California 96103 office phone: (805) 893-4072 e-mail: pikus@physics.ucsb.edu FAX : (805) 893-8422 ------------------------------------------------------------------------------- From: Mark Edwards Are you using PVM today? Yes. > What are your application(s)? I am part of team here at NIST involved in the creation of the NIST Parallel Applications Development Environment (PADE). This program employs a graphical user interface under X-windows to allow parallel applications developers using PVM to write complex parallel apps. Programmers can edit/compile/run and debug their PVM applications while staying within the same environment. More info about the PADE can be found on the World-Wide-Web at http://physics.nist.gov/ResOpp/hpcc/pade.html for even more info you can email me (Mark Edwards) at edwards@bruce.nist.gov We are considering a release in early Feb., 1995 -mark edwards ------------------------------------------------------------------------------- From: panwar@CS.UCLA.EDU (Ramesh Panwar) > Are you using PVM today? Yes. > What are your application(s)? Medical imaging: wavelet compression, image segmentation, image registration, volumetric image rendering, blood flow modeling etc. > How many and what kind of machines have you used with PVM? SUNs for developing test codes. UCLA Medical Center is in the process of purchasing a parallel machine. ------------------------------------------------------------------------------- From: Sinisa Veseli 1. Yes 2. Solving physics problems. 3. I've been using pvm with F90 on the NEXT machines (25 machines). There were several pvm bugs that I had to fix in order to make it work with F90 compiler. If you are interested, feel free to contact me. S. Veseli ------------------------------------------------------------------------------- From: Michael Rezny > Are you using PVM today? ANSWER: Yes I use pvm extensively. > What are your application(s)? ANSWER: Spatial modelling utilising a client - server approach. > How many and what kind of machines have you used with PVM? ANSWER: Workstations : Silicon Graphics ( 2), Sun (>20), DEC Alpha ( 3), Supercomputers : Cray YMP/2D ( 1), YMP/4E ( 1), MPP : Intel Paragon ( 1), IBM SP2 ( 1). regards Mike Rezny ------------------------------------------------------------------------------- From: David Chasman > > Are you using PVM today? yes. > > What are your application(s)? a platform independent scalable electronic structure code. > > How many and what kind of machines have you used with PVM? about 5 > > Any comments and suggestions for improvement are welcome. A shared counter would be VERY helpful. ------------------------------------------------------------------------------- From: Emilio Gallicchio Yes, I'm currently using PVM. My application is Monte Carlo simulation of quantum liquids. I'm using PVM locally on a network of IBM RISC6000 workstations for debugging purposes and remotely on a IBM SP2 machine for production runs (although I plan to implement MPL on this last machine). Emilio Gallicchio Department of Chemistry Columbia University ------------------------------------------------------------------------------- From: zdenko@harvey.cam.rice.edu (Zdenko Tomasic) >Are you using PVM today? yes. >What are your application(s)? large sparse symmetric matrix diagonalization >How many and what kind of machines have you used with PVM? sun, cray t3d >Any comments and suggestions for improvement are welcome. make sure when vendors claim to have implemented PVM to really implement everything and not just a convenient subset, e.g. t3d pvm fakes a dozen pvm calls returning error and considers pvmfadvise an unsatisfied external. ------------------------------------------------------------------------------- From: Xiaoguang Zhang ~{UEO~9b~} > Are you using PVM today? Yes. > > What are your application(s)? First principles calculations as well as small model calculations. > > How many and what kind of machines have you used with PVM? 10 IBM RS/6000 IBM SP2. > > Any comments and suggestions for improvement are welcome. > Please email the information to pvm-survey@cs.utk.edu. > Paths for input and output files should be specified at the console or in the hostfile. The current version puts the input and output files in the HOME directory and is very inconvenient. For example, in FORTRAN if unit 8 is written to without an open statement, a file 'fort.8' is created. It is very bad to put this in one's HOME directory. Many programs that are ported from serial versions will have this kind of write statements in them, and it would be helpful to be able to put them in a subdirectory specified by the user. ------------------------------------------------------------------------------- From: Anthony Fiarito >Are you using PVM today? YES. >What are your application(s)? POVRAY (Persistence of Vision Ray Tracer) >How many and what kind of machines have you used with PVM? 50+ Sun SPARCStations (SunOS 4.1.x & SunOS 5.4) 5 SGI Indy Workstations (Irix 5.2) /\nthony --- Anthony Fiarito alf@ee.pdx.edu Computer Action Team http://www.ee.pdx.edu/~alf Portland State University Chaos. Live it. Learn it. Be it. Project KPSU 1450AM - Least Significant Digit - Saturdays - 11pm to 12a ------------------------------------------------------------------------------- From: "Mr. Package" > Are you using PVM today? No. It turned out PVM was overkill for my small project. I plan on using PVM for some other work in the future, however. > What are your application(s)? Visual Recognition; exploring ATM networking. > How many and what kind of machines have you used with PVM? HP9000 7xx, between 5 and 15. I hope the survey results are useful, Kelly ------------------------------------------------------------------------------- From: debra@cs.unr.edu (Debra Estabrook) I am not currently using PVM. I used it in the past on fractals and used DECstations as my machines. ------------------------------------------------------------------------------- From: Jon A Webb I'm using PVM on an SP2, as the message-passing layer for an HPF-variant compiler (the CMU Fx compiler). -- J ------------------------------------------------------------------------------- From: symes@masc39.rice.edu (William Symes) > > Are you using PVM today? Yes. > > What are your application(s)? Inverse problems in seismology > > How many and what kind of machines have you used with PVM? 1) Ethernet connected cluster of IBM RS6K/370s 2) " " " " HP 735s 3) Cray C90 - pvm task distribution on top of autotasking, very effective on dedicated machine > > Any comments and suggestions for improvement are welcome. > Please email the information to pvm-survey@cs.utk.edu. > A problem in the past has been the necessity to use independent instrumentation (alog, upshot). xpvm may have taken care of this. will know when we have used it more extensively. Will pass this on to other members of my group for comment. ------------------------------------------------------------------------------- From: Susan Feingold 1. We are in the process of installing PVM. 2. Our applications are a mix of engineering and pure science codes. 3. We are currently installing PVM on an SMP 6 processor, shared memory DEC/APX Alpha 7000/660. We might also use it in a heterogeneous mix of Computer Center servers, including two Hyundai Axil 311/5.2's (Sun clones) and two RS/6000's, and of course the DEC Alpha. - ~~ - ~~ - ~~ - ~~ - ~~ - ~~ - ~~ - ~~ - ~~ - ~~ - ~~ - ~~ - ~~ - ~~ - ~~ - ~~ Susan Feingold D.Sc. ccasuzi@techunix.technion.ac.il Taub Computer Center phone 972-4-293696 Technion, Israel Institute of Technology fax 972-4-236212 Haifa, Israel 32000 - ~~ - ~~ - ~~ - ~~ - ~~ - ~~ - ~~ - ~~ - ~~ - ~~ - ~~ - ~~ - ~~ - ~~ - ~~ - ~~ ------------------------------------------------------------------------------- From: Raj Thiagarajan Yes I still use PVM Molecular Dynamics is my application Machines : SGI family mostly ------------------------------------------------------------------------------- From: Tal Yonatan hello Project Team, I am pleased to take place in your survey. 1. Yes i am using PVM today (version 3.2). 2. My applications is numerical simulation of turbulent pipe flow and it's post processing. 3. I am using 10 DEC 3000 (ALPHA) work stations in the Dep. of fluid mechanics in tel aviv university. each of my slaves programs is using about 50 Mbytes of memory. please contact me for more information. Jonathan Tal ------------------------------------------------------------------------------- From: Nicolas Droux > Are you using PVM today? YES > What are your application(s)? - PPC: Portable Parallel C (http://www.info.isbiel.ch/ppc/) - Parallel library implemented on top of PPC - ECS: External Computation System (an package allowing Mathematical applications (Mathematica, ...) to export numerical intensive calculations to parallel and distributed systems) >How many and what kind of machines have you used with PVM? 20 x SPARC 10, Solaris 2.4 3 x NEXTSTEP computers (and soon about 10 Linux machines) Best Regards, Nicolas Droux Engineering School of Biel-Bienne Switzerland ------------------------------------------------------------------------------- From: lks@ll.mit.edu (Keith Sisterson) Message-Id: <9501081004.AA16649@LL.MIT.EDU> To: pvm-survey@cs.utk.edu Subject: Re: PVM survey PVM Team, We are not using PVM at the moment. We are considering it for a new project which will ultimately require a combination of workstations and perhaps special processor boxes totaling well over 10 GFLOPS peak. The system is for handling large volumes of images and extracting information from them. We will soon be studying whether or not PVM is an appropriate tool for this job, and one question which springs to mind is that of the overhead through the PVM calls. We do have the PVM manual, but I have done little but glance at it yet, so if the answer is there I shall no doubt find it. If there is an e-mail address to which we can send such questions, I would be grateful if you let me know it. Thanks, Keith Sisterson (lks@ll.mit.edu) (617) 981 3465 ------------------------------------------------------------------------------- From: noe@olivia.quim.ucm.es (Noe Garcia Almarza) > > Are you using PVM today? Yes, I am, but I am still an unexperienced beginner > > What are your application(s)? Molecular Dynamics > > How many and what kind of machines have you used with PVM? IBM 6000, dec 5000 > > Any comments and suggestions for improvement are welcome. Best wishes, and thanks for producing PVM > The PVM Project Team Noe G. Almarza ------------------------------------------------------------------------------- From: "Rene' Kulschewski" > >Are you using PVM today? Yes. > >What are your application(s)? Numerical applications in chemical engineering. > >How many and what kind of machines have you used with PVM? 1 Sun running SunOS 4.1.3, 1 Sun running Solaris 2.3, 1 IBM RS 6K with AIX, 1 NeXTstation. > >Any comments and suggestions for improvement are welcome. I would like to see the ability of synchronous communication,e.g. task A get a notification that task B got the message from A (kind of 'blocking send'). It would be nice if one could specify priorities for communication between processes. I mean that i would like to say: message-passing between task A and B is prioritized over communication between B and C. I would find it useful if i could specify that the order between messages send is guaranteed ( following scenario: Task A send's messages to B,C,C,B; it would be nice if not B get his last message before C got his first ...). Regards Rene' --- ____________________________________________________________________________ Rene' Kulschewski ------------------------------------------------------------------------------- From: nprasad@eos.hitc.com (Narayan S. Prasad) >> >>Are you using PVM today? Yes >> >>What are your application(s)? Scientific >> >>How many and what kind of machines have you used with PVM? All major workstations like HP, SUN, SGI, IBM. Currently using it on the Cray T3D and IBM SP-2. >> >> >********************************************************* >Michael Bopf Phone:(301)925-0398 >Hughes Applied Information Systems Fax: (301)925-0326 >********************************************************* > > > *********************************************************************** Narayan S. Prasad PH: (301) 925-0467 Dept. SDPS FAX: (301) 925-0327 EOSDIS Core System Hughes Applied Information Systems, Inc. 1616 McCormick Drive Landover, MD 20785 *********************************************************************** ------------------------------------------------------------------------------- From: axon@cortex.rutgers.edu (Ralph Siegel) Are you using PVM today? At the moment I am using PVM3.3. What are your application(s)? Large scale simulations of collections of neurons in the visual cortex. Neural networks. How many and what kind of machines have you used with PVM? A ethernet cluster of 5 Rs/6000 Model 540, 370, 370, 370, powerpc Maui SP/2 50 to 100 nodes at once. -- Sincerely, Ralph Siegel, Ph.D. Assistant Professor axon@cortex.rutgers.edu phone: 201-648-1080 x3261 fax: 201-648-1272 Center for Molecular and Behavioral Neuroscience Rutgers, The State University 197 University Avenue Newark, NJ 07102 For additional information and reprints use Mosaic: http://www.cmbn.rutgers.edu/cmbn/faculty/rsiegel.html ------------------------------------------------------------------------------- From: KARG > > Are you using PVM today? > yes > What are your application(s)? > Parallel FFT, parallel solution of differential equations > How many and what kind of machines have you used with PVM? > 4 HP-UX machines 2 models 700 and 2 models 800 Mfg. -------------------------------------------------------------------------- | /\ /\ +----------------------------------+ | / \ / \ |\/|ichael | Dipl.-Ing. Michael Karg | | / \/ \ +----------------------------------+ | /\ |/ | 0222/58801/5420 od. 0222/5056190 | | / \ |\arg | karg@uranus.tuwien.ac.at | | /----\ | Graf Starhemberggasse 32/2/3 | |/ \ +---+ 1040 Wien +----+ | http:/uranus.tuwien.ac.at/~karg/karg.html | +-------------------------------------------+ ------------------------------------------------------------------------------- From: F.A.Rabhi@computer-science.hull.ac.uk (Fethi A Rabhi) > >Are you using PVM today? > Yes. >What are your application(s)? > PVM is used as a target language for high level parallel programming environment. We are also using PVM for some Monte Carlo simulation and molecular dynamics in collaboration with the Applied Physics Department. >How many and what kind of machines have you used with PVM? > Network of Sun Sparcs. 8-node i860 based Transtech Paramid. --------------------------------------------------------------------- Dr Fethi A. Rabhi Email : F.A.Rabhi@dcs.hull.ac.uk Computer Science University of Hull Tel : +44 (0)1482 465744 Hull HU6 7RX (UK) Fax : +44 (0)1482 466666 ------------------------------------------------------------------------------- From: Hiroki Konaka > Are you using PVM today? Yes, I'm using PVM 3.2.6. > What are your application(s)? I'm implementing a parallel object-oriented language, called OCore. I use PVM as the communication infrastructure of the OCore runtime system for NOW's. > How many and what kind of machines have you used with PVM? Sun SPARC stations. Thanks for your efforts to provide such a good environment. - Hiroki Konaka ------------------------------------------------------------------------------- From: cxprc@rc.rit.edu (Chris Pane) > > Are you using PVM today? Yes I am > > What are your application(s)? > I am using pvm to implement distributed rendering applications (Raytracing at the moment). This particular project is part of my Master Thesis. > How many and what kind of machines have you used with PVM? 50 SUN SPARC II 25 SGI5 Machines 4 Pentium (Running Linux) Looking into a Sparc station with Transputers on board (Not sure if this is possible, is it ?????). 2 SPARC 10 machines ------------------------------------------------------------------------------- From: Toshiyuki Takahashi $@!d(JAre you using PVM today? Yes. $@!d(JWhat are your application(s)? KLIC $@!d(JHow many and what kind of machines have you used with PVM? Sun SS10 * 6 IBM RS6000/25T * 6 ---- # Toshiyuki TAKAHASHI # Takeda Lab. # Dept. of Information Science # Science Univ. of Tokyo # E-mail:tosiyuki@is.noda.sut.ac.jp ------------------------------------------------------------------------------- From: rsweet@tiger.denver.colorado.edu (Roland Sweet) I am currently not using PVM although I plan to use it within the next nine months to create a distributed memory parallel version of the software package Crayfishpak. Roland Sweet ------------------------------------------------------------------------------- From: rory@chemical-eng.edinburgh.ac.uk 1) Yes, I am using PVM today. 2) I have and am using PVM for 2 things: (i) As one means of parallelising a Dynamic Chemical Process Simulator. I previously used Meiko CS-Tools, but wanted to expand out into the use of work station clusters. (ii) As the communication and program management layer for a distributed, object based engineering environment we are developing. PVM is used to manage the machines being used, spawn new object manipulation applications (methods) and to pass ascii encoded objects between them. It is a Client/Server based environment were applications are continually being dynamically spawed and killed. We also have applications joining sessions without ever being spawned within PVM. 3) So far we have used SUN4s, a LINUX box and IBM RS6000s. I have also tried to use a SEQUENT running DYNIX/ptx(R) V2.1.1 but could not get it to compile. The functions gettimeofday() and ffs() were not available or no BSD compatibility library appears to exist for this OS. On System V machines the function get_clock seems to be the alternative to gettimeofday(). Unfortunately there is no allowance in the PVM source for easily substituting gettimeofday as there is for other functions such as bcopy() for example. Comments: --------- Overall, I think PVM is an excellent package. My main complaints are as follows. (i) Output management: I really hate the way output is managed. You either have to tail -f a /tmp file which usually hangs when too much output appears or you have to manage all of the output youself in an application you have specified wants it using the pvm_setopt function. The second of these is fine in principle. It would however be a nice touch to add an option which tells PVM to manage the incoming redirected output for the programmer. i.e. Whenever PVM gets a message which is an output redirect is simply dumps it to stdout/stderr rather than expecting the programmer to deal with it. (ii) pvm_kill() One of our projects uses PVM to spawn and manage many applications. Part of this management process is to kill application which have hung or are nolonger wanted. It is not always possible to send a shutdown message, and quite often we have to use pvm_kill(). The problem is that pvm_kill() only sends a SIGTERM, which some applications ignore. I would like to see pvm_kill() extended to allow SIGKILL as well as SIGTERM, which is guaranteed to kill the application. In fact, could pvm_kill() not be expanded to behave in the same manner as the unix kill(), thus allowing a multitude of signals to be sent to distributed applications. At the moment we have to log into the machine and kill the application via a shell which is inconvenient and outwith the applications control. (iii) PVM to PVM connection. PVM only allows applications which belong to a single user to communicate. It would be a usefull extension to allow one users PVM to log in to and send messages to another users PVM. In our design application each user works under a client/server environment. However we want each user to be able to use other users object data bases and applications. To do this we need to connect the servers together. At present we are having to use sockets since PVM does not support inter PVM communication at present. Hope all this helps Rory. ------------------------------------------------------------------------------- From: Oren Laden Hi, PVM survey >>Are you using PVM today? Yes. >>What are your application(s)? We are using PVM as a reference execution environment to test our dynamic load-balancing mechanisms. Specific issues include load distribution and redistribution, high utilization of a NOW configuration, and communication overhead. We are also running parallel numerical applications. >>How many and what kind of machines have you used with PVM? A cluster of Pentiums, 486/66 and 486/33 PC's. >>Any comments and suggestions for improvement ... The current work and load distribution of PVM should be improved. We executed PVM on top of MOSIX (our NOW OS) and observed an improvement of 25-150% in the execution times of several parallel applications. These results are expected to be even worse in a multiuser usage. Bye, Oren. **************************************** ** Oren Laden (orenl@cs.huji.ac.il) ** ** ---------------------------------- ** ** Distributed Operating Systems lab. ** ** The Hebrew University of Jerusalem ** ** Jerusalem, Israel 91904 ** **************************************** ------------------------------------------------------------------------------- From: christian.leder@stud.uni-rostock.de (Christian Leder) > > Are you using PVM today? no, not today but on a regular basis approx. 2 times per month > What are your application(s)? numerical computations, simulations, work on my thesis > How many and what kind of machines have you used with PVM? SUN cluster, RS6000 cluster, LINUX PC cluster ------------------------------------------------------------------------------- From: Carlos Armando Magalhaes Duarte > > Are you using PVM today? > Yes! > What are your application(s)? > Parallel implementation of hp Finite element Methods. > How many and what kind of machines have you used with PVM? > It is been developed on a cluster of 16 RS6000 (many models). > Any comments and suggestions for improvement are welcome. Keep up with the good work. ------------------------------------------------------------------------------- From: "Paul Hayes" > > Are you using PVM today? Not at this moment, but I plan to. > > What are your application(s)? Large memory matrices. Sets of large memory matrices. > > How many and what kind of machines have you used with PVM? Sun, HP, alpha. > > Any comments and suggestions for improvement are welcome. > Please email the information to pvm-survey@cs.utk.edu. > More example programs would be nice. Perhaps the fellows who developed lparx could get some good programs/utilities for you. -- Paul Hayes Special Purpose Processor Development Group Phone:(507) 284 3479 Mayo Foundation Fax:(507) 284 9171 Guggenheim 1012B Email:hayes.paul@mayo.edu 200 First Street SW Rochester, MN 55905 ------------------------------------------------------------------------------- From: Julian Hall > Are you using PVM today? > Yes > What are your application(s)? > Parallel revised simplex method for linear programming. > How many and what kind of machines have you used with PVM? > Local network of SUN workstations and the Edinburgh Cray T3D. ------------------------------------------------------------------------------- From: rainer@parchaos.mathematik.uni-freiburg.de (Rainer Kleinrensing) Hello there, > Are you using PVM today? Yes, we use it on our 14 SGI Indigo Workstations and on a Sun Sparcstation Pool. > What are your application(s)? We are doing CFD and have ported some code to a transputer machine (Parsytec Xplorer, NOT using PVM). We want to be able to port it to a variety of machines, so we consider changing it to use PVM. There is a commercial implementation of PVM on the Parsytec machine, but we did not use this one yet. > How many and what kind of machines have you used with PVM Only the SGIs and Suns mentioned, but we'll try it on the Parsytec machine soon. It's a nice piece of software, thanks a lot ! -- Rainer Kleinrensing, rainer@mathematik.uni-freiburg.de Institute for Applied Mathematics Phone: +49-761-203-5639 Hermann-Herder-Str. 10 Freiburg University Fax: +49-761-203-5632 79104 Freiburg / Brsg., Germany ------------------------------------------------------------------------------- From: louis@ford.ee.up.ac.za (Louis Coetzee) To: pvm-survey@cs.utk.edu Subject: Re: PVM survey >>>>> On Fri, 6 Jan 1995 18:08:47 -0500, pvm-survey@cs.utk.edu said: PVM> Are you using PVM today? No, not today, although I use it very often. PVM> What are your application(s)? Parallel training of neural networks as well as a variety of parallel optimization algorithms. PVM> How many and what kind of machines have you used with PVM? 12 IBM RS6000 machines or 1 4 processor SUN sparcserver 1000 PVM> Any comments and suggestions for improvement are welcome. I would really appreciate a shared memory port for the KSR1. Cheers Louis Coetzee louis@ford.ee.up.ac.za ------------------------------------------------------------------------------- From: Ralf Hetzel Q11: I am not using PVM at the moment. Q2: Quantum Monte Carlo simulations in theoretical solid state physics Q3: About 12 HP workstations, 700 series, running HP-UX 8.07 Best regards, Ralf Hetzel ------------------------------------------------------------------------------- From: tom (Thomas Heiling) > > Are you using PVM today? Yes > > What are your application(s)? A parallel chess program > > How many and what kind of machines have you used with PVM? My main machines are 386 or 486 PC with varying Speed and equipment. Normally the program plays on 386/25 MHZ 4 MB RAM 386/40 MHZ 8 MB RAM 486/33 MHZ 16 MB RAM 486/66 MHZ 8 MB RAM and eventually on some HP Workstations for testing purpose. ( HP9000/ 715 ) ------------------------------------------------------------------------------- From: atwood@nas.nasa.gov (Christopher A. Atwood) > > Are you using PVM today? y > > What are your application(s)? computational fluids, nonlinear dynamics, controls > > How many and what kind of machines have you used with PVM? v3.2.3: 19 nodes, w/ machine composed of: sgi (mips r3000, r4000,r4400), hp755 (pa-risc), ibm590 (rios) pvme on ibm sp2 to 31 nodes ------------------------------------------------------------------------------- From: Karl.F.Moschner@urlus.sprint.com About a year ago we downloaded a copy of PVM and successfully tested it but did not move ahead with implementing critical applications under PVM. It was attractive but a few factors kept us from pursuing PVM: - we are low on manpower - commercial applications offered greater speedups on a single workstations than we would have attained by implementing comparable available codes under PVM - we purchased a parallel server which supported all of our software without PVM and offered greater performance than all of our workstations combined We have encouraged suppliers to offer PVM variants but have had no success. One particular commercial vendor plans to offer coarse grained network parallelism (automatically submit a series of jobs or readily discernable subprocesses across a series of identified accounts/systems). Karl F. Moschner Karl.F.Moschner@urlus.sprint.com Unilever Research U. S. ------------------------------------------------------------------------------- From: John Chitalia 1) Are you using PVM today? No, but I plan to use it when I implement my algorithm. 2) What are your application(s)? Parallel graphics algorithms. Mainly geometric modelling. 3) How many and what kind of machines have you used with PVM? I've used a network of Sun workstations (up to 10 different ones) and an Intel iPSC/2 with 4 and 16 nodes. Since I haven't actually programmed with PVM yet, I can't offer any suggestions. I've only tried out the test examples. Sorry about that. JC chitalia@pixel.csee.usf.edu ------------------------------------------------------------------------------- From: Mike Kupferschmid > Are you using PVM today? Not this precise day, but I have a graduate student who is using it from time to time these days. > What are your application(s)? He's trying to use parallel function evaluations to speed up a nonlinear optimization algorithm. > How many and what kind of machines have you used with PVM? We've been running on a 16-node IBM SP1, whose nodes are RS6000/360 workstations. > Any comments and suggestions for improvement are welcome. We have been hindered in our attempts to devise an effective parallel algorithm by the fact that PVM does not provide a primitive for asynchronously interrupting a worker process and causing it to start over again as though it had been killed and respawned. Instead we must either kill and respawn the process (which has a very high overhead indeed) or poll within the worker (which is inconvenient and also wastes time). There should be a way within the worker to set a program interrupt exit: "whenever you get this signal from the master program, stop whatever it is you are doing and go back to statement n". And there should be a way within the master to send the signal that says "abandon that calculation and get ready for a new assignment". ------------------------------------------------------------------------------- From: zliu@tiger.denver.colorado.edu (Zhining Liu) Are you using PVM today? Yes, but not very much What are your application(s)? CFD and numerical combustion How many and what kind of machines have you used with PVM? IBM-SP2 Zhining Liu ------------------------------------------------------------------------------- From: MAVERICK Are you using PVM today? Yes. What are your application(s)? Scientific data (images and times series) analysis/processing. How many and what kind of machines have you used with PVM? 6 Decstations. adios Ata <(|)>. | Mail Dr Ata Etemadi, Blackett Laboratory, | | Space and Atmospheric Physics Group, | | Imperial College of Science, Technology, and Medicine, | | Janet atae@uk.ac.ic.ph.spva Earn/Bitnet atae@spva.ph.ic.ac.uk | | Internet/Arpanet atae%spva.ph.ic.ac.uk Span SPVA::atae | ------------------------------------------------------------------------------- From: Ed_Donley Response to PVM Survey 1. Are you using PVM today? I am using PVM to teach an undergraduate course on Numerical Methods for Supercomputers. The title is now a misnomer, since I am using PVM in addition to the Pittsburgh Supercomputing Center's Cray YPM C90. 2. What are your application(s)? The students are using the Lennard-Jones model (a system of ordinary differential equations) for molecular dynamics modelling. This unit is based on Lloyd Fosdick's and Liz Jessup's High Performance Scientific Computing course materials at the University of Colorado. 3. How many and what kind of machines have you used with PVM? Six DECstation 5000s running Ultrix 4.4. Ed Donley Mathematics Department Indiana University of PA Indiana, PA 15705 U.S.A. ------------------------------------------------------------------------------- From: --> --> Are you using PVM today? Yes, but only just started. --> --> What are your application(s)? Distributed process support --> --> How many and what kind of machines have you used with PVM? Currently 5 SUN4 (SunOS 4.1.3) machines -- Stewart Brodie Dept. Electronics & Computer Science, Southampton University, UK. http://louis.ecs.soton.ac.uk/~snb94r/ ------------------------------------------------------------------------------- From: Are you using PVM today? Yes. What are your application(s)? Large-scale parallel simulation of models in statistical mechanics. How many and what kinds of machines have you used with PVM? Clarkson University RISC/6000 cluster: approx. 70 IBM RISC/6000 workstations. SUN Workstations at Los Alamos National Laboratory: approx. 20 SPARCstations. Suggestions for improvement: It would be very, very nice to be able to transfer "unsigned" variables between processes (i.e. in C, unsigned long, unsigned int, etc...). In the version of PVM I have now, it is only possible to transfer signed long's and int's. ------------------------------------------------------------------------------- From: Himavan P 1. Yes, I am using PVM. 2. Application: Finite Element Analysis, Stiffness Matrix Generation & Assemblage. 3. How many machines/what kind: 2 to 4 machines, Sparc Stations SUN. x. Thank you, Himavan Pamujula ------------------------------------------------------------------------------- From: Sameer Suresh Shende > > Are you using PVM today? No. I am a CS grad student and I was using it last term for course projects. > > What are your application(s)? > Expanding the PVM API. Writing system software. I am not an application developer. > How many and what kind of machines have you used with PVM? About 6 SUN4 workstations (SunOS) and a Sequent Symmetry machine (Dynix). > > Any comments and suggestions for improvement are welcome. I was working on a dynamic load balancing API for PVM. PVM currently provides point to point communication constructs for communication (or multicast). The sender iss aware of who is going to receive the message. I developed an API based on heirarchical farms where the farmer task makes a number of woprk packets and sends it to a central server. The workers group themselves into process worker classes (multiple farms per application, multiple worker classes per farm, multiple worker tasks (pvm tasks) per worker class, worker can in turn be a farmer - heirarchical farm, and dynamic worker task migration from one farm to another) and receive work packets only when they request for one. This way, the workers on slow nodes get fewer work packets and the application is load balanced. If you want the report of my project, I will be glad to give it to you. My address is sameer@cs.uoregon.edu ------------------------------------------------------------------------------- From: Steve Moyer >Are you using PVM today? Yes. >What are your application(s)? I am using PVM as a communications and process management substrate for parallel-distributed file system research (PIOUS). >How many and what kind of machines have you used with PVM? 1) Sun4/SunOS 4.1.3 2) Sun4/SunOS 5.3 3) SGI/IRIX 4.0.5 4) Dec Alpha/OSF1 2.1 5) HP 9000/HP-UX 6) IBM RS6000/AIX >Any comments and suggestions for improvement are welcome. Two suggestions: 1) I would like to see the bug fixed that prevents pvm_pkbyte()/pvm_upkbyte() pairs from being able to pack/unpack in different groupings; eg 6 calls to pvm_pkbyte() packing one byte each and 2 calls to pvm_upkbyte() unpacking 3 bytes each. This works in the data raw mode but not in the data default mode. Since this can be done with all other types, it should be made to function properly with the byte type. I have reported this before, but was told that this is considered a feature and not a bug. 2) Message contexts for library development would be a BIG improvement. Regards, Steve ------------------------------------------------------------------------- Steven A. Moyer Email: moyer@mathcs.emory.edu Mathematics and Computer Science Office: (404) 727-0668 Emory University, Atlanta, GA 30322 Fax: (404) 727-5611 ------------------------------------------------------------------------------- From: kly@sunpaw.cern.ch (Gregory KOZLOVSKY) >>Are you using PVM today? No, I finished a program which uses PVM and this program will hopefully be used by other people. >>How many and what kind of machines have you used with PVM? I used PVM on SP-1 and SP-2. >>Any comments and suggestions for improvement are welcome. I would like to see better debugging facilities. There should be a way of getting the output from different host on your terminal (prefixed with host id) and not into a file only as it is now. ------------------------------------------------------------------------------- From: bbs@bbsipc.newcastle.edu.au (Bryan Beresford-Smith) Are you using PVM today? I used it for some of the practical work last year in a course I taught on Parallel Processing. What are your application(s)? So far, only as a tool for teaching and research. How many and what kind of machines have you used with PVM? About 20 SUN workstations and also with a MasPar MP-1. Any comments and suggestions for improvement are welcome. It has been very useful as a teaching tool. Thanks! ****************************************************************************** Dr Bryan Beresford-Smith Senior Lecturer .-_|\ Dept of Computer Science email: bbs@cs.newcastle.edu.au / \ The University of Newcastle ph. + 61 49 216056 \.--._* <- Callaghan NSW 2308 fax + 61 49 216929 v Australia ****************************************************************************** ------------------------------------------------------------------------------- From: nhan@tiny.me.su.oz.au (Nhan Phan-Thien) > Are you using PVM today ... Yes, we are planning to expand our PVM farm > What are the applications ... Our applications are in the areas of Stokes flow and elasticity using boundary element methods. Mainly the operator is deflated and an iterative technique is used to obtain the solution. The system matrix is not stored but calculated as needed. A sub-domain decomposition is used, mapping each sub-domain to each processor. Extension of the finive volume method to PVM is being planned. > How many machines ... We're using 3 DEC alpha 300 1 DEC alpha 400 3 DEC alpha 800 2 Titan Ardent Ref: N Phan-Thien & D.L. Tullock, ``Completed Double Layer BEM in Elasticity and Stokes Flow: Distributed Computing Through PVM'', Computational Mechanics, 14 (1994) 370--383. Nhan Phan-Thien Professor in Mechanical Engineering ------------------------------------------------------------------------------- From: lange@cis.ksu.edu (Gregory Lange) > > Are you using PVM today? YES! > > What are your application(s)? > Running a monte carlo simulation involving turbulent flow fields. > How many and what kind of machines have you used with PVM? > 5 Sparc10's an am now running on the NCSA power challenge. -- Gregory Lange lange@cis.ksu.edu ------------------------------------------------------------------------------- From: Paul Crivelli > Are you using PVM today? Yes, I have been since around November. > What are your application(s)? I have parabolized a Fluid Mechanics program (originally written by someone else). I am comparing their in-house message passing protocol to PVM. Sorry but I cant give you any more info than that. Confidentiality and all that dont ya know. > How many and what kind of machines have you used with PVM? All DEC machines. Predominantly Alphas (2-4) but also a few of their older machines (1-2). The documentation helped me a great deal. Especially the Appendix covering all the different commands, this was invaluable. The quick reference card helped alot too. Congratulations on a job well done. Im a Mechanical Engineer and I didnt have a lot of problems with learning PVM. Paul Crivelli ------------------------------------------------------------------------------- From: Jason Moore > Are you using PVM today? Yes. > > What are your application(s)? Mainly research. Parallel I/O, getting parameters for models. > > How many and what kind of machines have you used with PVM? Usually no more than 8 machines: HPPA, SUN4 and Meiko CS-2 using their modified version of PVM as well as the daemon-based generic version ------- Jason Moore Internet: moorej@research.cs.orst.edu Department of Computer Science Oregon State University Bell-net: (503) 737-4052 "Anything worth doing is worth doing with a smile" ------------------------------------------------------------------------------- From: bneta@math.nps.navy.mil (Beny Neta) Sirs Yes I am using PVM currently on a network of sun workstations (up to 16 machines). Application: Orbit Prediction. Beny Neta ------------------------------------------------------------------------------- From: Prachya Chalermwat > > Are you using PVM today? YES. > > What are your application(s)? NOT YET DECIDED BUT TEND TO BE IMAGE PROCESSING, MATRIX COMPUTATION. > > How many and what kind of machines have you used with PVM? 2 Linux-based 486 machines 1 Sparc 1000 server 2 Sun servers 2 Sun workstations 1 Sun Sparc workstation > -- PC =-------------------------------+-----------------------------------= Prachya Chalermwat | "I've tried everything. email: prachya@seas.gwu.edu | I've not failed. [O] (202) 994-5268 | I've just found 10,000 ways [Fax] (202) 994-5296 | that won't work." | Thomas A. Edison =-------------------------------+-----------------------------------= ------------------------------------------------------------------------------- From: cstokley@cs.gmu.edu (Katja Stokley) Are you using PVM today? Not specifically today (it's Sunday :-)), but yes, I'm using PVM now. What are your application(s)? Simulating an interconnection network for a massively parallel computer in order to implement algorithms for the simulated machines. How many and what kind of machines have you used with PVM? Sparc workstations, up to about 20 of them. Sincerely, Katja Stokley cstokley@cs.gmu.edu ------------------------------------------------------------------------------- From: Jonathan Kua > > Are you using PVM today? > Yes. > What are your application(s)? We were trying to do execution time prediction on-the-fly so as to be able to schedule a parallelizable program as efficiently as possible. We concentrated on medium-grain parallelism. > > How many and what kind of machines have you used with PVM? > HP 715s Sun Sparc 10s IBM RS/6000 We have only used heterogeneous combinations of up till 20 machines recently. We might try more later. Jonathan Kua kua@crhc.uiuc.edu ------------------------------------------------------------------------------- From: tsune@taklab.takenaka.co.jp (TSUNEKAWA Hiroshi) Dear Sir: pvm-survey>Are you using PVM today? Yes pvm-survey>What are your application(s)? I'm using PVM to parallelize EDEM: Extended Distinct Element Method that is based on Cundall's DEM and that have beed developed by Prof. Hakuno. It is a method to analyze both discrete and continuous material. pvm-survey>How many and what kind of machines have you used with PVM? I usually use two or three SGI machines. Indigo2(IRIX5) CRIMSON(IRIX5...recently changed...I haven't excuted PVM yet) 310VGX(IRIX4) EDEM needs more traffic data than FEM. I'm trying to reduce the data to accelerate the execution. sincerely yours. ----- TSUNEKAWA Hiroshi Fundamental Research Department TAKENAKA Research & Development Institute, Chiba, Japan ------------------------------------------------------------------------------- From: roberto@ee.uwa.edu.au (Roberto Togneri) > > Are you using PVM today? Not exactly today, but we have been using it off and on and it will be used for parallelising some algorithms we have on workstation clusters. There is currently a vacation student learning PVM with the intent to use it for something before the summer is out (see below). > > What are your application(s)? We have used PVM as a vehcile for a load sharing program we developed and I am still hoping we can incorporate it as part of PVM. Currently we are looking at using it to parallelise the Expectation-Maximisation (EM) algorithm used in our work. It is a prime candidate due to its computational and memory requirements. > > How many and what kind of machines have you used with PVM? So far just Suns (SPARC II, ELCs) running SunOS 4.1.3. We have a two-processor SGI Onyx and if a native message-passing version of PVM is available (I haven't checked the latest version yet) it may used to explore/exploit the SGI. > > Any comments and suggestions for improvement are welcome. I have a project looking at the various ways and means of communicating using the TCP protocols directly to examine the most effective communication standard for parallel algorithms (e.g connectionless UDP .vs. connectioned streams). One problem I found with PVM was that its internal buffering meant that if I wanted to transer X Mb of data I would need to have 2X Mb of memory. I haven't yet tested the latest version which allows user control of buffer management. On the otherhand for most parallel algorithms there are many small-data (< 1024 kb) communications and a way to optimally handle this would be useful (this is what the above project is going to try and figure out). If you would like further clarification or have suggestions please do not hesitate to contact me. I am definitely interested in the area of parallel computation over networked computers. Regards, -- Dr. Roberto Togneri Phone: +61-9-380-2535 _--_|\ Centre for Intelligent Information Processing Systems / \ Dept. of Electrical & Electronic Engineering *_.--._/ The University of Western Australia Fax: +61-9-380-1101 v NEDLANDS WA 6907 Australia Email: roberto@ee.uwa.edu.au ------------------------------------------------------------------------------- From: sato@mt.cssi.co.jp (Ryuuichi Sato) Please receive my answers for your three questions. 1. Yes, I am using PVM now. 2. My application is Fluid Flow Analysis / FDM. 3. The machines I use are two or four SUN4s. I hope these help you. Best regards, Ryuichi SATO, Canon Supercomputing S.I., Osaka JAPAN, e-mail: sato@cssi.co.jp ------------------------------------------------------------------------------- From: Here's my answer, >> Are you using PVM today? >> Yes. >> What are your application(s)? >> CFD(Computational Fluid Dynamics). >> How many and what kind of machines have you used with PVM? >> approximately four to ten of networked IBM's RS6000. I'm considering of using a workstation cluster of 16 nodes in the future. Regards. Takashi Ohta at ohtak@trl.ibm.co.jp -------------------------------------------- Tokyo Research Laboratory / IBM Japan ------------------------------------------------------------------------------- From: Hiroshi NAKASHIMA > Are you using PVM today? Well, although I'm not sure whether I'll use PVM today, I and my coleagues are using it almost daily. > What are your application(s)? Implementation of a concurrent logic programming language. PVM is the inter-processor communication kernel of the run-time system of the language processor. > How many and what kind of machines have you used with PVM? Two. Fujitsu AP1000, a message-passing-base distributed-memory multicomputer, and SunSparc workstations connected by LAN. ----------------------------------------------------------------- Hiroshi Nakashima (nakasima@kuis.kyoto-u.ac.jp) Dept. of Information Science, Faculty of Engineering, Kyoto Univ. Yoshida Hon-Machi, Sakyo-Ku, Kyoto, 606-01 Japan. +75-753-5383 information to pvm-survey@cs.utk.edu. I might ask you to add a configuration for EWS4800s in the next distribution. I will send you one when I confirm that the test suite runs well. Best regards, Koichi Konishi Computer System Lab., C&C Labs. NEC Corporation Phone: +81-44-856-8488, FAX: +81-44-856-2231 ------------------------------------------------------------------------------- From: zhang hao > Are you using PVM today? Yes, we use it. > What are your application(s)? We use it to do some research about WS. Cluster. And to present a platform such that the user of our MPP machine can run their Applications on it. They do have a lot of Apps. :-) > How many and what kind of machines have you used with PVM? I haven't use a lot of nodes to run PVM. But I did have run it on a lot kind of platforms, includes: SUN4, RS6K, MIPS, E88K, Motorola S-900 and Mach 2.6 running on PC 386. Hope that helps to improve PVM --Zhang Hao NCIC, China. ------------------------------------------------------------------------------- From: hideaki@hydra.cray.com (Hideaki Moriyama) > > Are you using PVM today? > NOT today. Once a month, maybe. > What are your application(s)? > I don't have real applications. Just play with it. > How many and what kind of machines have you used with PVM? > Cray (Vector, MPP and CS6400) DEC 10000 IBM 590 Sun (Solaris 1 and 2) HP SGI > Any comments and suggestions for improvement are welcome. > Please email the information to pvm-survey@cs.utk.edu. > compatibility between versions ------------------------------------------------------------------------------- From: pplab@myrinae.hallym.ac.kr (Project PVM) Hi, Thanks for your efforts. 1.Yes, I am using PVM today. 2.Parallel numerical processing & Program analysis 3. 8 (SUN3, SUN4) Have a nice day (-| jkjang@sun.hallym.ac.kr pplab@maple.hallym.ac.kr ------------------------------------------------------------------------------- From: Suresh Hungenahally Hi I am using PVM extensively for teaching and research. I use it to teach in the Advanced Computer Systems Course at the Honours level in my School and ususlly I get 22 to 24 post graduate students in this subject. I also use it for applied projects in Signal processing and other algorithmic work. In fact I am finding it very hard to process the data i am getting just due to mere volume. I would appreciate if any one can collaborate with me to get the reports converted into journal papers. I am willing to offer the source code and the co-authorship. At present I have 14 technical reports sitting on my desk all of them are work using PVM on 8 machines. reg. suresh _______________________________________________________________________________ Suresh Hungenahally suresh@tomahawk.sct.gu.edu.au Signal Processing and Intelligent Systems Research Group, School of Microelectronic Engineering, Faculty of Science and Technology, Griffith University, Nathan Qld 4111 Australia Ph +61 7 875 5063 Fax +61 7 875 5198 ------------------------------------------------------------------------------- From: David Sitsky > Are you using PVM today? Yes - although most of my programs are now being written in MPI. > > What are your application(s)? General Relativistic ray tracer. > > How many and what kind of machines have you used with PVM? Development under a SUN farm, runs on the Fujitsu AP1000 machine at the ANU and CM5 machines at the ANU, university of New South Wales and University of Adelaide. A run was done combining the three CM5s and the AP1000. David Sitsky Email: sits@cs.anu.edu.au CAP Research Program Phone: +61 6 249 5143 Australian National University Fax: +61 6 249 0010 Canberra ACT 0200, Australia WWW: ftp://dcssoft.anu.edu.au/pub/www/dcs/people/sitd/sits.html ------------------------------------------------------------------------------- From: Craig Huckabee > > Are you using PVM today? No, not as of yet. I have not used it since school (See Question #3) but I'd hope to find a use for it at my new job. > What are your application(s)? > > How many and what kind of machines have you used with PVM? I used PVM as a senior at Clemson University for a class project. We were assigned the task of using PVM to do a distributed merge-sort between 5 or more machines (simple task, but it was used to demonstrate process control and communication for the class which was about Operating Systems) I had been a cooperative education student with Naval Electronic Systems Engineering Center, (NAVELEX), Charleston - a DOD/Navy command, and had access to several Sun machines that were distributed across the US. I used PVM to do my merge sort between 5 machines, each being in a different city in the US. Each machine was an identical Sun SPARC 10/41 and the links were T1 and ethernet. I got a 'A' for the effort and I found PVM to be easy to use and well documented. I guess I should kinda say Thanks! > Regards, Craig Huckabee NAVELEX Charleston, SC huck@nosc.mil ------------------------------------------------------------------------------- From: chik@icot.or.jp (Takashi Chikayama) Hideaki Moriyama (hideaki@hydra.cray.com) forwarded me your message calling for feedback on PVM. Here's mine. Are you using PVM today? Yes. What are your application(s)? In an implementation of a parallel logic programming language called KLIC (available by anonymous ftp from ftp.icot.or.jp). There are several other (more efficient) implementations dependent on platforms, but the PVM version is meant to be portable. How many and what kind of machines have you used with PVM? - SparcCenter 2000 (20 processors) - DEC AXP-7000 (6 processors) - SparcStation 10/52 systems, connected with ethernet (8 systems with 16 processors in total) Comments: To build a symbolic processing system for dynamic and irregular computation, it is desirable to asynchronously interrupt other processes to stop or suspend speculative computation and start computation newly known to be mandatory. We use sendsig for the purpose, but sendsig sometimes outstrips ordinary messages, which is a problem because the receiver may receive no messages after receiving the signal. Our implemetation currently use timer-driven polling in combination to make it up, but it is an awkward solution. It would be much better if PVM could provide some feature that guarantees that the signal arrives *after* specified messages (though I'm not sure whether it can be implemented in a portable manner or not). Takashi Chikayama Institute for New Generation Computer Technology (aka ICOT) Tokyo, Japan ------------------------------------------------------------------------------- From: "Wang Dao-liu" I am learning to use PVM now and plan to apply it to Molecular Dynamics. Now we have installed PVM on 8 SGI indigo R4000 connected in a local network and some i486s, which are used for simulation in single machine. We have found PVM is quite good environment for network computation. Many thanks for your supporting PVM. Best regards, Daoliu Wang ------------------------------------------------------------------------------- From: Shinji Hioki >> >> Are you using PVM today? >> Yes, but only test for production. >> What are your application(s)? >> Elementary Particle Physics Lattice Quantum Chromo Dynamics Monte Carlo Simulation >> How many and what kind of machines have you used with PVM? >> 4 DEC ALPHA S.Hioki Dept. of Physics, Hiroshima Univ., Higashi-Hiroshima 724, JAPAN ------------------------------------------------------------------------------- From: sjchen@pine.iecs.fcu.edu.tw (Shih-Jeh Chen) Dear pvm_survey: Thank you for giving the important message. I am using PVM today! My applications are parallel processing research, and MPI standard implementation. I use four or eight SUN-workstations(SPARC classic) for PVM applications. I need the informations about PVM and MPI, if you have the lastest message, please mail to me. Thank you very much! email: sjchen@pine.iecs.fcu.edu.tw Name : Shih-Jey Chen ------------------------------------------------------------------------------- From : Feng-Chia University, Taichung, Taiwan. > >Are you using PVM today? No, not today. I used it a year ago and I will probably use PVM again in a few months. >What are your application(s)? Experimental meso-scale wind model. >How many and what kind of machines have you used with PVM? 1 x HP735/99 1 x HP715/50 1 x HP705 1 x SPARC 2 1 x SPARC ELC 2 x SPARC 1 1 x SPARC SLC Regards, Peter McGavin. ------------------------------------------------------------------------------- From: lally@lanl.gov (Bryan Lally) > >Are you using PVM today? No. I hope to be soon. My codes are almost ready for this step. >What are your application(s)? CFD, always coupled with some other transport phenomena (heat transfer, electromagnetics, particulate flow, etc.) and often with chemcial reactions layered on top of the fluid mechanics). >How many and what kind of machines have you used with PVM? None, but hope to use a heterogeneous cluster of SGI and DEC AXP workstations. - Bryan -------------- Bryan Lally lally@lanl.gov ------------------------------------------------------------------------------- From: Tony von Sadovszky Are you using PVM today? Yes Iam using PVM today.... What are your application(s)? My applications include the following: - particle in cell plasma modeling - genetic algorithms applied to ground water modeling, eclipsing binary star modeling, and electron scattering resonance calculations How many and what kind of machines have you used with PVM? The machines upon which I utilize PVM include: - 23 DECstation 3100's - 4 HP 9000/725's - IBM SP2 @ Muai - CRAY T3D @ Anchorage -- ------------------------------------------------------------------------------ Tony von Sadovszky Email: tony@rigel.physics.unr.edu University of Nevada, Reno Office (LP 314): (702) 784 - 6059 Department of Physics/220 Department: (702) 784 - 6792 Reno, Nevada 89557-0058 Fax: (702) 784 - 1398 ------------------------------------------------------------------------------ From: muramatu@cx_nkkc34.info-sys.tokyo.nkk.co.jp (muramatu) Dear the person in charge, This is Muramatsu at NKK Co., Japan. In reply to the message from you. >Are you using PVM today? No. I used to use PVM. >What are your application(s)? An ab initio Molecular Orbital Method programme. >How many and what kind of machines have you used with PVM? On CONVEX C34, HP9000 4cpu cluster. Sorry that I cannot do much help. I moved to TCGMSG because the programme I'm using was revised to fit TCGMSG. I think PVM is much more polular than TCGMSG, and is the de fact standard. Besides this, in the field of chemistry, TCGMSG has something. Regards, -------------------------------------------------------- Yukihisa Muramatsu: ------------------------------------------------------------------------------- From: nomura@exp.cl.nec.co.jp >Are you using PVM today? Yes, I am. >What are your application(s)? Kinds of basic utilities for computer simulations. >How many and what kind of machines have you used with PVM? I using NEC EWS4800 workstation claster: usually 6 to 10. >Any comments and suggestions for improvement are welcome. >Please email the information to pvm-survey@cs.utk.edu. I wish to use "numerial recipes for C++ using PVM"!! ********************************************************** Masahide Nomura (nomura@exp.cl.nec.co.jp) ------------------------------------------------------------------------------- From: mansha@research.trddc.ernet.in (Rajesh K. Mansharamani) Are you using PVM today? Yes What are your application(s)? CFD, Sintering (Engineering simulations) How many and what kind of machines have you used with PVM? SUN Sparc II, SUN3, RS6K, I386, I486 (ATT SYS V and LINUX), HP9000/720, HP9000/435 Any comments and suggestions for improvement are welcome. Please email the information to pvm-survey@cs.utk.edu. PVM is extremely portable and reasonably efficient. We wish the packing calls could be automated. For example, have a parser that parses a function call such as pvm_send(tid, tag, arg1, arg2, ...) where arg1, arg2 can be of different types (e.g, complex structures) and by determining the types of the arguments it generates pack calls. This could be a preprocessor to C compilation. Also if something can be provided under PVM for the advanced system programmer who might want to use various socket options (such as select etc.) it will be useful. Rajesh K. Mansharamani TRDDC Email: mansha@trddc.ernet.in 1 Mangaldas Road Phone: (212) 622 809 Pune 411 001 Fax: (212) 623 713 India ------------------------------------------------------------------------------- From: pefv774@hpcf.cc.utexas.edu (Taskin Ucpinar) Here are the answers to your questions. > > Are you using PVM today? > Yes, very extensivelly. > > What are your application(s)? > My program is the extension to the utchem code of the program that is developped by the University of Texas @ Austin. > > How many and what kind of machines have you used with PVM? > The program has been run under a heterogeneous network of work stations of a total of 12 machines. The network consists of DEC-ALPHA, IBM-RS6000 and IBM-PowerPC computers. On the other hand, a modified version of the same program runs under Cray-T3D. > > Any comments and suggestions for improvement are welcome. > I wish it was a little bit faster on a heterogenous network. For comparison a version of utchem runs under c-linda system. And that version is faster than the pvm version. I know that the information that I gave is not detailed. If you need more information please do not hesitate to contact. Taskin Ucpinar University of Texas @ Austin Petroleum and Geosystem Eng. Dept. ------------------------------------------------------------------------------- From: tchen@elbert.Mines.Colorado.EDU (Tong Chen) Are you using PVM today? Yes. What are your application(s)? Imaging the earth's subsurface structure for oil and gas exploration. How many and what kind of machines have you used with PVM? I've used about 20 IBM RS6000 and 10 NeXT workstations, mainly for experiment. Normally, I only use less than 10 IBM RS6000. ------------------------------------------------------------------------------- From: naiman@super.org (Aaron E. Naiman) Hi. Although I am not currently using PVM, I hope to use it in the future. In the past I have set it up, and used it (slightly) for a CM-5. Thank you. -- Aaron Naiman | IDA/SRC | University of Maryland, Dept. of Mathematics | naiman@super.org | naiman@math.umd.edu ------------------------------------------------------------------------------- From: scott stanley > > Are you using PVM today? I am not currently using PVM very much. However, I will be using it in the future. I am working with a vector code and will be using PVM later to help develop a parallel version of the code. > > What are your application(s)? My applications are in the area of computational fluid dynamics (both direct numerical simulation and large eddy simulation of turbulent flows.) > > How many and what kind of machines have you used with PVM? 12 HP 9000 series workstations ------------------------------------------------------------------------------- From: Eddie Johnson Hello, I am not currently using PVM today, but I did use it in a class on parallel computing. We used PVM on 15 Sun Stations to implement a matrix multiplication routine and a prime number generator. The PVM platform ran very well, and I was quite impressed. One thing I noticed was that entering a large number of machines from the PVM command prompt to form a conference was sometimes cumbersome, and I wish there was a faster method of handling this. I wrote my own code to add machines to the conference, but it would be nice if this code was already in place, possibly in the form of reading a text file of host names and adding them to the conference? Just an idea... I would like to congratulate you on the design of such as amazing platform. Keep up the good work! Eddie Johnson -- __^__ __^__ ( _____ )-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-( _____ ) | | | | | | | | | | Eddie Johnson | | | | Undergraduate, Computer Science | | | | The University of North Carolina at Greensboro | | | | E-mail : johnsone@hamlet.uncg.edu | | | | | | | ___ | | ___ | ( _____ )-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-( _____ ) V V ------------------------------------------------------------------------------- From: > Are you using PVM today? Yes. > What are your application(s)? I have a program, Surface Evolver, that is an interactive program for modelling liquid surfaces. I have experimented with making a parallel version with PVM, and I plan more extensive experiments this spring. Also, next year I will be teaching a course on parallel programming, and I am thinking of using PVM there. Do you know of any course materials based on PVM? > How many and what kind of machines have you used with PVM? SGI, Sun, NeXT, both solo and network of 10-15 mixed types. Also a little on a CM5. And solo on an Intel machine. Ken Brakke brakke@geom.umn.edu ------------------------------------------------------------------------------- From: loebel@ZIB-Berlin.DE (Andreas Loebel) Dear PVM Project Team, thank you for your interest in my current work, >Are you using PVM today? > yes I do, at this time I have installed pvm 3.3.5 >What are your application(s)? > We delevop a distributed branch&cut program for vehicle scheduling in urban public mass transit. The distribution on serveral workstations is possible and necessary because we are interested in solving a kind of multi-commodity-flow problem with up to 30 million variables. The communication with pvm is not very extensive. >How many and what kind of machines have you used with PVM? > At this time we use a cluster of SUN workstations (sparc20, sparc10, sparc5). The operation system is solaris 2.3. >Any comments and suggestions for improvement are welcome. >Please email the information to pvm-survey@cs.utk.edu. > For us pvm is ok. The only problem we sometimes have is adding pvm daemons from the pvm console. Best wishes, -- Andreas Loebel Konrad-Zuse-Zentrum fuer Informationstechnik Berlin (ZIB) Heilbronner Str. 10, D-10711 Berlin (Wilmersdorf) PHONE: +49 30 / 89604-239 EMAIL: loebel@zib-berlin.de ------------------------------------------------------------------------------- From: hilka@em2c.ecp.fr (Martin Hilka) 1/ Yes, We are still using PVM 2/ Our Applications are principally Direct Numerical Simulation of Reactive Flow (turbulent premixed and diffusion flames). - CFD in a wider sense. 3/ Machines used are Sun (Sparc10), SGI and RS6000 clusters, T3D, CS2, CM-5, Paragon, SP-1/SP-2 and VPP-500. For production runs we target Workstation clusters, T3D and VPP-500, perhaps SP-2. Thanks to you all. Martin Hilka ------------------------------------------------------------------------------- From: Thomas Frank > Are you using PVM today? Yes ! > What are your application(s)? CFD - Computational Fluid Dynamics, Multiphase Flows > How many and what kind of machines have you used with PVM? 3 machines HP 735/755 , MIMD parallel Computer Parsytec GC with up to 128 Motorola Power PC Chips > Any comments and suggestions for improvement are welcome. > Please email the information to pvm-survey@cs.utk.edu. It would be fine to have in PVM the same mechanism for message induced processes like in EXPRESS. In EXPRESS it is possible to define a special category of messages. If such a message is send to a curtain processor, a user-defined process is initiated. In that user-defined procedure it is possible to send further messages (normal and special type messages). So you can submitt results of that procedure call to the initiating processor. That kind of special messages gives you a convenient tool for programming of - for instance - memory managers for large amounts of distributed data. Sincerely yours, Thomas Frank ******************************************************************************* * Dr. Thomas Frank | * * Technical University Chemnitz-Zwickau | Adress of the Institute : * * Faculty of Mechanical Engineering and | * * Process Technology | * * Department of Multiphase Flow | Reichenhainer Strasze 88 * * PSF 964 , 09009 Chemnitz | 09126 Chemnitz * * Germany | Germany * *-----------------------------------------------------------------------------* * Tel.: +49 (371) 531 46 43 * * Fax.: +49 (371) 531 46 44 email : Frank@IMech.TU-Chemnitz.DE * * URL : http://www.tu-chemnitz.de/~fth/mpf/mpf_home.html * ******************************************************************************* ------------------------------------------------------------------------------- From: jleik@statoil.no (Joergen Leiknes) > >Are you using PVM today? Yes > >What are your application(s)? ECLIPSE, a reservoir simulation program. > >How many and what kind of machines have you used with PVM? Approx. 10 machines, IBM-rs6000, HP-hppa, SGI and sun4. > ------------------------------------------------------------------------------- From: "Dr. Manfred Kunicke" >Are you using PVM today? Yes we do, mostly in a test mode to get some experience in programming and controlling (XPVM). >What are your application(s)? currently there are no "production applications" >How many and what kind of machines have you used with PVM? HP9000(2*735, 1*715) regards m.kunicke ------------------------------------------------------------------------------- From: sbokhari@shb.lhe.imran.pk (Shahid H. Bokhari) > Are you using PVM today? Yes > What are your application(s)? Parallel computing research and teaching. We have just started and expect to have some specific applications on soon. At that point we will be in a position to offer specific suggestions for improvement. > How many and what kind of machines have you used with PVM? 4 IBM PC Compatible 486/66 +2 386/33, running Linux. Thanks for PVM! Shahid Bokhari Department of Electrical Engineering University of Engineering & Technology Lahore 54890 Pakistan ------------------------------------------------------------------------------- From: Gerhard Wilhelms > Are you using PVM today? Yes. > > What are your application(s)? PVM is used as educational tool for lectures in parallel programming. > > How many and what kind of machines have you used with PVM? ~30 IBM RS/6000 [lines deleted] Best regards, Gerhard Wilhelms _____________________________________________________________________ University of Augsburg (Germany) Department of Mathematics - Computer Science Universitaetsstrasse 2 Phone : +49 +821/598-2176 D-86159 Augsburg Fax : 598-2200 Internet: Wilhelms@Informatik.Uni-Augsburg.DE ------------------------------------------------------------------------------- From: alm01@sican.de (Praktikant AR(W.van Almsick)) > > Are you using PVM today? > Yes, I do. > What are your application(s)? > I'm on myself just making benchmarks on diferent (virtual) parallel architectures. We have written a "benchmark generator" so everyone can make his own benchmarks on his own parallel architecture with his own application specific workload. So everyone who wants to use a (virtual) parallel machine can test the computing and the communication behaviour before he buys the machine. > How many and what kind of machines have you used with PVM? > I'm using up to eight HPPA workstations, a CONVEX parallel computer and in future a CRAY. > Any comments and suggestions for improvement are welcome. > Please email the information to pvm-survey@cs.utk.edu. > We like to run our project on an NCUBE because it grows up on an it under PICL. But there is no version of PVM for the NCUBE. So, we would like to have an version for the NCUBE. I hope this helps you, Karsten Zander ************************************************************* * * * * Karsten Zander * E-Mail: alm01@sican.de * * * zander@irb.uni-hannover.de * * * * * * Germany, University of Hannover * * * * ************************************************************* ------------------------------------------------------------------------------- From: jan@nats2.informatik.uni-hamburg.de (Jan Amtrup) > Are you using PVM today? Yes. > What are your application(s)? PVM we implemented a communications facility modeled after the channels of Occam and the transputer. We designed interfaces for C, C++, Common Lisp, Prolog and Tcl/Tk. The communication system is actually used within the German joint research project Verbmobil for message passing between different modules. Additionaly, it is used to explore novel architectonic issues for natural language understanding systems, which are inverstigated as a subproject of Verbmobil. > How many and what kind of machines have you used with PVM? I personally use Sun Sparstations 2 and 10 under SunOs 4.1.x and Solaris 2.3 Other partners in our project use HP Workstations under HPUX or IBM RS6000 (and maybe some others which I don't know of). Regards, Jan. _______________________________________________________________________ Jan Willers Amtrup Phone: (+49 40) 54 715-519 Universitaet Hamburg Fax: (+49 40) 54 715-515 FB Informatik (AB NatS) e-mail: amtrup@informatik.uni-hamburg.de Vogt-Koelln-Str. 30 jan@nats2.informatik.uni-hamburg.de D-22527 Hamburg GERMANY _______________________________________________________________________ ------------------------------------------------------------------------------- From: Winfrid Tschiedel > > Are you using PVM today? Yes > > What are your application(s)? So we are selling computers, our main interest is, that PVM is available on these platforms. We are just using PVM in benchmarks > > How many and what kind of machines have you used with PVM? Fujitsu M780/10 running UXP/M V12L10 several SGI Workstations running IRIX 5.2 SGI Powerchallenge (4 x R8000 ) IRIX 6.0 KSR-1 ( 16 processor ) KSR OS 1.2.2 -- _______________________________________________________________________________ Siemens Nixdorf Informationssysteme AG SH WI Email: winfrid.tschiedel@mch.sni.de Otto-Hahn-Ring 6 Tel. : ++49-89-636-45652 81739 Muenchen Fax : ++49-89-636-45046 ------------------------------------------------------------------------------- From: Renate Wahl > > Are you using PVM today? I have installed it and I use it for tests and for courses of lectures. And I will be happy if I have waken the interest in it. > > What are your application(s)? I know that a student experiment with PVM on many ULTRIX - machines and in the medici- nal computer center PVM is installed also, but I don't know, what they do. They have AIX - machines, I think. > > How many and what kind of machines have you used with PVM? In our central computer center I have installed PVM at 1 ALPHA AXP 500 2 DEC Station 5100 1 IBM RISC 560, 340, 320 and 5 X 220 Best wishes, Renate Wahl -+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+- I E - MAIL : zrw@hpux.rz.uni-jena.de I Renate Wahl I I Phone : ( +49 03641) 636290 I Rechenzentrum der FSU Jena I I FAX : ( +49 03641) 636354 I Am Johannisfriedhof 2 I I I 07743 Jena I I I Germany I -+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+- ------------------------------------------------------------------------------- From: mseritan@edil.dcae.pub.ro (Marius Seritan) > Are you using PVM today? Yes, I am using PVM. I have installed the library for more than one year but I have started to use it only 2-3 month ago. > What are your application(s)? I am working in the field of semiconductor devices modelling. I am using PVM in order to implement a distributed simulator. I am interested by distributed numerical computation algorithms. > How many and what kind of machines have you used with PVM? PVM is installed on 6 IBM RS6000 machines running AIX 3.2.{1,3,5}. Yours sincerely, Marius Seritan Marius Seritan http://edil.dcae.pub.ro:8001/~mseritan Department of Electronics and Telecomunications Polytechnica University of Bucharest Romania ------------------------------------------------------------------------------- From: Yes, I'm using PVM at the moment. I'm using PVM in the development of a parallel regularized decomposition algorithm for the solution of large scale nonlinear stochastic programming problems. Subproblems are solved by GAMS, called as a child process from within the PVM processes. I'm using PVM on a network of SUN workstations running SUNOS and SOLARIS. Regards, Charles H. Rosa (rosa@iiasa.ac.at) ------------------------------------------------------------------------------- From: > Are you using PVM today? Yes. > What are your application(s)? Implementation of parallelizing/distributing compilers. > How many and what kind of machines have you used with PVM? 1. Network of 30 DEC-Alpha 2. Network of 60 Sun-Sparc 3. Cray-T3D with 256 processors > Any comments and suggestions for improvement are welcome. 1. I would like to have for instance better support for a logical task numbering scheme, instead of having to assign the numbers myself at program initialization from the array of TIDs given to pvm_spawn. Program initialization in general is a bit heavy to my taste. 2. Why not include some kind of support for RPC ? It seems to me that this also is a mechanism many users have to implement manually in PVM. Best regards J.Hulaas, hulaas@di.epfl.ch Compiler Lab, Swiss Federal Institute of Technology, Lausanne ------------------------------------------------------------------------------- From: Bernd Kugelmann > > Are you using PVM today? Yes > > What are your application(s)? Numerical analysis --- boundary value problems --- numerical solution of ode > > How many and what kind of machines have you used with PVM? 20 --- SUN SPARCs --- SGIs (connected by ETHERNET) -- +-----------------------------------------------------------------------------+ | Dr. Bernd Kugelmann Phone: Germany + 89/2105-8128 | | Fax: Germany + 89/2105-8156 | | E-mail: kugelm@mathematik.tu-muenchen.de | | Mathematisches Institut | | Technische Universitaet | | D-80290 Muenchen | | Germany | | | +-----------------------------------------------------------------------------+ ------------------------------------------------------------------------------- From: thoeny@ito.umnw.ethz.ch (Juerg Thoeny ETHZ) Dear makers of PVM, first of all, I would like to thank you for publishing PVM. >Are you using PVM today? Yes. >What are your application(s)? I'm using PVM to run simulations in parallel on our network. The simulations are time-continous, time-discrete and event-driven. >How many and what kind of machines have you used with PVM? I'm using four SUN Workstations (with a total of five processors). >Any comments and suggestions for improvement are welcome. >Please email the information to pvm-survey@cs.utk.edu. I was up and running in a short time. The only problem was my own mistake. I put the PVM_ROOT environment variable into the interactive section of my .cshrc file. I discussed this in the news-groups. It seems, that I'm not the only person, which did this wrong. It could be helpfull to put a warning into the manual. Kind regards Juerg ''' (o o) +---oOO--( )--OOo------------------------------------------------+--------+ !Juerg Thoeny, Systems Ecology, Institute of Terrestrial Ecology,! 2b ! !Grabenstrasse 3, CH-8952 Schlieren/Zuerich, Switzerland ! or ! !tel +41 1 633-6092, e-mail thoeny@ito.umnw.ethz.ch ! not 2b ! +--oOO---------OOo-----------------------------------------------+--------+ ------------------------------------------------------------------------------- From: Gerd Meister Hi, > Are you using PVM today? Yes. > > What are your application(s)? > I am working on Parallel Discrete Event Simulation and have implemented a parallel logic simulator based on VHDL. I am especially interested in the possibilities for speeding up sequential simulations using "parallel" machines. > How many and what kind of machines have you used with PVM? > I am using 4 to 30 SUN4's running either Solaris 2.3 or BSD 4.1.3, an Intel iPSC/860 Hypercube and 2 SUNMPs (Sparc Station 20 with 2 processors each). Bye, Gerd. -- Gerd Meister Tel.: 06151/16-5411 TH Darmstadt FB Informatik Fax : 06151/16-4070 FG Verteilte Systeme, Prof. Mattern D-64293 Darmstadt Email : meister@isa.informatik.th-darmstadt.de ------------------------------------------------------------------------------- ------------------------------------------------------------------------------- From: ae906am@s600.rz.rwth-aachen.de (Herr_an_Mey) Dear PVM Project Team, thanks for your question. We installed PVM 3.2 on a Fujitsu VP2600 at our site and tried an simplified implementation ov PVM 2 on the Fujitsu VPP 500 vector parallel computer. But as we received a PARMACS implementation for our computers, which delivers more performance on the VPP500, we do not use PVM for production purposes at the moment. Because we have such a large mono processor, the users do not need to parallelize their applications. In future this situation might change. Probably our interest might turn to MPI than. I will forward your question to my collegue Stefan Karsch, who is more involved in pVM for academical reasons. Best wishes Dieter an Mey (user support for vector and vector parallel applications) -------------------------------------------------------------------- \_\_\_\_ \_ \_\_\_\_\_ \_ Dieter an Mey \_ \_\_ \__ \_ \_ \_ \_ RWTH Aachen Rechenzentrum \_\_\_ \_\_\_\_ \_ \_\_\_ Seffenter Weg 23 D-52074 Aachen \_ \_ \__ \__ \_ \_ \_ phone: +49(0)241-804377 fax: -8888134 \_ \_ \_ \_ \_ \_ \_ e-mail: anmey@rz.rwth-aachen.de -------------------------------------------------------------------- ------------------------------------------------------------------------------- From: vernard@cc.gatech.edu (Vernard C. Martin) > Are you using PVM today? yes > What are your application(s)? Monitoring of distributed applications > How many and what kind of machines have you used with PVM? I use PVM on several different architectures. Sun, SGI, KSR, IBM and Cray. I tend to use about 8 machines for the workstations and 1 64 node KSR and a 4 processor CRAY. > Any comments and suggestions for improvement are welcome. > Please email the information to pvm-survey@cs.utk.edu. Not at the moment. -- Vernard Martin Vernard Martin (vernard@cc.gatech.edu) Georgia Tech College of Computing "To absent friends, lost loves, old gods, and the season of mists; and may each and every one of us always give the devil his due." ------------------------------------------------------------------------------- From: john.doyle@cen.jrc.it (John Piero Doyle) 1) No, I am not using PVM today. 2) I am working on a Monte Carlo fortran code to simulate the radiative transfer of photons in an ocean-atmosphere system. 3) I am presently working on the code and have not yet implemented the parallelization of the computation of the photon histories on multiple processors, but have the firm intention of doing so in the future. To do this I was thinking of using PVM. 4) Can you give me an anonymous ftp adress (or something similar) where I can find the latest version of PVM? Many thanks, John ---------------------------------------------------------------------------- John Piero Doyle T.P. 272 Unit Marine Environment (ME) Institute for Remote Sensing Applications (IRSA) Joint Research Centre (JRC) European Commission (EC) I-21020 ISPRA (VA) ITALY Direct tel: ++39 (0)332 789876 Fax: ++39 (0)332 789034 E-mail (INTERNET): john.doyle@cen.jrc.it ----------------------------------------------------------------------------- From: Dear Colleague, Present application of PVM is numerical solution of certain partial differential equations on cluster of workstations. Architectures used in tests: CNVX, HPPA, PMAX, RS6K, SUN4. Number of nodes in network: usually 1, but sometimes up to 10 or more. Regards, J.N. Buur. ------------------------------------------------------------------------------- From: Jacques.de.Swart@cwi.nl Dear Colleague, I am not using PVM already. But I want to use it for numerically solving ODEs in parallel. The machines on which I want to run PVM are clusters of workstations and Parsytec. Best wishes, Jacques de Swart Centre for Mathematics and Computer Science Amsterdam The Netherlands ------------------------------------------------------------------------------- From: Garf Bowen > > Are you using PVM today? Yes. > > What are your application(s)? A number of applications in and around a commercial oil-reservoir simulator. 1. Parallel version of the code. 2. co-operation between a number of simulator models. 3. Communication with 'controlling' graphics programs. > > How many and what kind of machines have you used with PVM? > Of the order of 20. Workstations including Sun/SGI/RS6000/HP/Dec. > Any comments and suggestions for improvement are welcome. > Please email the information to pvm-survey@cs.utk.edu. > I like the system. I've had some problems with non-blocking receives, but haven't checked them out with the latest versions. ------------------------------------------------------------------------------- From: Andreas Listl > Are you using PVM today? Yes. > > What are your application(s)? Parallel Database System. > > How many and what kind of machines have you used with PVM? 2-10 SUN SPARC and SUN 10 > Comments: We would like to have support for shared memory in PVM. In our application we use UNIX shared memory. We would also like to have support for threads as some kind of PVM-task. Best wishes, Andy. ------------------------------------------------------------------------------- From: Angela Gelli > > Are you using PVM today? No. > > What are your application(s)? > I used PVM to develop a parallel Branch and Bound code and some numerical prallel algorithms. > How many and what kind of machines have you used with PVM? > about 5: some HP720, HP730, HP735, SUN10, RISC6000 Ibm...and SP1 Ibm. Usually not all together... Best regards Angela Gelli ------------------------------------------------------------------------------- From: Klaus Wechsler Dear PVM Project Team, of course, I'd like to answer your questions: I will use PVM today. Currently, I intend to change from 3.2.6 to 3.3.6. I work on Computational Fluid Dynamics. Until now, I worked with SGI and SUN workstation cluster, but only with a limited number of machines (4-5). Thank you very much for your interest. Klaus Wechsler ------------------------------------------------------------------------------- From: garda@ief-paris-sud.fr (Patrick Garda) Are you using PVM today? YES What are your application(s)? Writing and running parallel programs for neural networks, image processing, simulated annealing and Monte-Carlo simulation of electronic devices. How many and what kind of machines have you used with PVM? 6 SUN workstations with Ethernet, 16 DEC ALPHA farm with GIGASWITCH. We plan to use it on an IBM SP1 with 8 processors in 1995. Any comments and suggestions for improvement are welcome. Please email the information to pvm-survey@cs.utk.edu. GO ON, We need you. Your help is much appreciated. ------------------------------------------------------------------------------- From: radtke@imbe05.imbe.uni-bremen.de (Stefan Radtke) > Are you using PVM today? YES > > What are your application(s)? Parallel test pattern generation for combinatorial, digital circuits. > How many and what kind of machines have you used with PVM? > About 1 to 20 SUN workstations from sparc 1+ to sparc 10 > Any comments and suggestions for improvement are welcome. Nice tool. Seems to become standard ! Furhtermore your response time for questions concerning problems is superb ! (I asked 2 or three times and Al Geist answered at the same day, some hours later) Thank YOU guys ! +----------------------------------------------------------------------------+ | Stefan Radtke Univ. of Bremen | | Dept.of Microelectronics| | E-Mail: radtke@imbe.uni-bremen.de P.O. Box 330 440 | | Voice: ++49 421 218-4439 D-28334 Bremen | | Fax: ++49 421 218-4318 Germany | | | | Home: | | Voice: ++49 421 616 7626 | | Fax: ++49 421 616 7627 | +----------------------------------------------------------------------------+ ------------------------------------------------------------------------------- From: kraft@busch.bauwesen.uni-dortmund.de (Manfred Krafczyk) > Are you using PVM today? > We are about to examine to use PVM as an alternative to EXPRESS which shows some instabilities when using many nodes (>15) in our workstation cluster. We have just installed XPvm and ran succesfully some 'hello_world' tests, but the main work is still to do. > What are your application(s)? > We are doing Fluid-Flow simulations with Lattice-Gas methods and Traffic Flow simulations with Cellular Automata, beside we are planning to do Finite Element parallelization. > How many and what kind of machines have you used with PVM? > We did the tests with up to 35 machines HP 715-50 and 735-100 (connected by standard ethernet) and the hello_worlds were o.k., beside we just received a Parsytec cluster with 16 nodes based on the Power-PC chip but this machine is being installed right now and there are no results yet. > Any comments and suggestions for improvement are welcome. > Having used EXPRESS in the last years I very much appreciated functions like automatic domain decomposition (mapping a matrix onto a cluster of nodes) and utility functions that effectively help sending parts of rows and columns of matrices through the system effectively. As these are tasks that almost everybody needs is there something in progress or even completed yet ? It would be a waste of time if everybody hacks these things on his own ?! Best regards Manfred Krafczyk -- **************************************************** * Dipl.Phys. Manfred Krafczyk * * Lehrstuhl NMI * (Numerische Methoden und Informationsverarbeitung) * FB Bauwesen, Universitaet Dortmund * * August-Schmidt-Str. 8 * 44221 Dortmund GERMANY * * * Tel.: ++49 231 755 2093 * Fax : ++49 231 755 2532 * e-mail: kraft@busch.bauwesen.uni-dortmund.de * * **************************************************** ------------------------------------------------------------------------------- From: fares@host11.lctn.u-nancy.fr (M. Fares) Yes I used PVM My application is the computational chemistry M. Fares ------------------------------------------------------------------------------- From: Marcus Naraidoo > > Are you using PVM today? Yes > > What are your application(s)? Applications are written in house to perform computational stealth calculations which can be parallelised to run on a variety of platforms, such as SG Challenge and CRAY T3D machines. Efficiency on such platforms is high. > > How many and what kind of machines have you used with PVM? > Only a network of DEC alpha machines. > Any comments and suggestions for improvement are welcome. It would be useful if, upon taking a node out of the network, all of the results didn't become jibberish. The probability of node removal increases with the number of available nodes and this appears to be the situation when PVM would be most useful. ------------------------------------------------------------------------------- From: Morten Hermanrud > Are you using PVM today? We are using PVM on our supercomputer cluster where I am system architect and administrator. > What are your application(s)? Our users are mainly scientific numeric calculations. > How many and what kind of machines have you used with PVM? We have 16 IBM RS/6000 model 590 and a 16 node SP2. Our main problem with PVM (from a system administrator viewpoint) is the many incompatible versions that needs to be maintained. ----------------------------------------------------------------------- Morten Hermanrud, (Morten.Hermanrud@ibm.no) AIX System Engineer, IBM Nordic Scientific and Technical Solutions #include ------------------------------------------------------------------------------- From: gallego@tcomhp20.epfl.ch (Cesar Jose Gallego) > Are you using PVM today? Not for the moment but we certainly will. We are currently installing an ATM Local Area Network in our Lab based on the Fore Systems' switch and their adaptation cards for workstations. As I saw that a PVM version that works with the Fore System API exists, I'm waiting our LAN to be ready to do the first tests with PVM over ATM. > > What are your application(s)? Distributed simulations of ATM switches and networks. > How many and what kind of machines have you used with PVM? Our Lab. currently has about 15 HP workstations and 3 SUNs. 8 of these stations will be connected through our ATM switch. The others are connected to the Ethernet backbone. THIS MAIL HAS BEEN SENT IN 100% RECYCLED BANDWIDTH. ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ ~ Cesar Jose Gallego ~ ~ Ecole Polytechnique Federale de Lausanne ~ ~ Laboratoire de Telecommunications ~ ~ 1015 LAUSANNE ~ ~ SWITZERLAND ~ ~ ~ ~ E-mail : gallego@tcom.epfl.ch ~ ~ Phone : (+41 21) 693 46 93 ~ ~ Fax : (+41 21) 693 26 83 ~ ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ ------------------------------------------------------------------------------- From: I'm using PVM, release 3.3.3, currently being upgraded. The applications running in PVM environment are linear algebra routines and Finite-Difference Time-Domain codes (for the analysis of electromagnetic propagation in wave-guides). At the moment, I have clustered 2 HP ws (715 and 735) and an IBM 320h. I'm going to add an IBM 250t also. A porting of other applications, including some combinatorial optimization codes using Tabu Search methods and Finite-Difference Frequaency-Dependent Time-Domain codes, is planned in the next future under PVM. Two main fields can be identified for our use of PVM codes: 1) Microwave wave-guides engineering 2) Effects of Electromagnetic radiations on biological systems That's all. Bye bye and happy new year to everyone! Luciano Tarricone Dept. of Electronics Universita' di Perugia, S. Lucia-Canetola, 06131, Perugia, Italy tel: 0039-75-5852663 fax: 0039-75-5852654 e-mail: tarricone@istel.ing.unipg.it ------------------------------------------------------------------------------- From: nlangloh@etro1.vub.ac.be (Nils Langloh) > > Are you using PVM today? Not this week, but we are planning to use it on a regular basis from the 23th January on. > What are your application(s)? We are developing a simulator for parallel optical computer architectures (optsim). > How many and what kind of machines have you used with PVM? Simultaneously: 12 machines: 7 SparcStation 20 running SunOs 4.1.3 1 SparcStation 10 with 2 processors running SunOs 4.1.3 1 SparcStation 2 running SunOs 4.1.3 1 SparcStation 1+ running SunOs 4.1.3 1 SparcStation 20 with 2 processors running SunOs 5.3 1 SparcStation 10 running SunOs 5.3 Best wishes, Nils Langloh ------------------------------------------------------------------------------- From: wsanpv@win.tue.nl (P.W.C. Vosbeek) > >Are you using PVM today? No, I used it only a few months ago. > >What are your application(s)? I am working an a method (Contour Dynamics) for simulating vortices in two dimensional flows of inviscid, incompressible fluids. > >How many and what kind of machines have you used with PVM? I used PVM with SUN workstations. The number of machines I used vary from 3 to 16. I am planning to work also with PVM on a Sillicon Graphics Power Challenge. > >Any comments and suggestions for improvement are welcome. Regards, Pauline Vosbeek ------------------------ Pauline Vosbeek wsanpv@win.tue.nl ------------------------------------------------------------------------------- From: >Are you using PVM today? Yes. >What are your application(s)? I use it for teaching purposes in the context of a module on parallel computing. The students can experiment with various parallel algorithms including matrix multiply, adaptive quadrature, merge sort, distributed control algorithm (mutual exclusion, deadlock detection, etc), travelling salesman problem. I also use it for research purposes (as a plateform for building a DSM system). We also use it occasionnally for student projects. >How many and what kind of machines have you used with PVM? I have used it with up to 10 SUN4/SUN SPARC workstations. >Any comments and suggestions for improvement are welcome. >Please email the information to pvm-survey@cs.utk.edu. Keep it simple and well documented. ID ------------------------------------------------------------------- | Isabelle Demeure Tel : 33-1-45-81-72-86 | | Departement Informatique Fax : 33-1-45-81-31-19 | | TELECOM Paris Email : demeure@inf.enst.fr| | 46, rue Barrault | | 75634 Paris cedex 13 | ------------------------------------------------------------------- ------------------------------------------------------------------------------- From: nimg@letterbox.rl.ac.uk (Nick Gould) In answer to your questions: ? Are you using PVM today ? Not today, but recently. ? What are your application(s) ? Two kinds: Genetic algorithms for large-scale optimization. The function evaluations, for our application, being performed in parallel. As these are by far the dominant cost, good speedups are possible. Solution of linear systems using both direct and iterative methods. Parallel multifrontal methods for the direct solution, and element-by- element preconditioned cg for the iterative solution. ? How many and what kind of machines have you used with PVM ? At one time I had a network running between the UK, France and Canada. This involved: 6 Suns, 2 RS/6000s, 6 Alphas, 5 HP/9000s and a YMP in the UK, 8 Suns and 4 RS/6000s in France and 2 Suns in Canada Hope this is useful, Nick ------------------------------------------------------------------------------- From: brozzis@falstaff (Stefano Brozzi) > > Are you using PVM today? Yes, for about 8hrs. > > What are your application(s)? I'm programming in PVM environment. I'm using XPVM and HeNCE. > > How many and what kind of machines have you used with PVM? > From a min. of 2 to a max of 10 or 12. 8 SUN4, 4 SUN Solaris > Any comments and suggestions for improvement are welcome. no comments for now. Brozzi Stefano brozzis@verdi.eng.unipr.it -- ------------------------------------------------------------------------------- From: "Philippe Besse" > Are you using PVM today? I'll use it next year. > What are your application(s)? statistical simulation (resampling as bootstrap) and may be stochastic algorithm (simulated annealing, genetic) > > How many and what kind of machines have you used with PVM? > ? -- o o o -----------------------------o ------------------------------------------ | Philippe C. Besse || ||||||||| CNRS U.A. D0745 | | Laboratoire de |||| o o || Universit'e Paul Sabatier | | Statistique et || |||| |||| 118, route de Narbonne | | Probabilit'es || |||| 31062 Toulouse cedex | | Email: besse@cict.fr ||||||| ||| France | | Tel: (33)61.55.61.42 Fax: (33)61.55.60.89 | ------------------------------------------------------------------------- ------------------------------------------------------------------------------- From: Riccardo Valente > Are you using PVM today? Yes (IBM's PVM/6000 version, though). > What are your application(s)? Solid State calculations, based on Density Functional Theory, Pseudopotentials and Plane Wave Expansion. Lots of communications. Speedup ~6 w/ 8-node SP1 > How many and what kind of machines have you used with PVM? IBM RS6000/580 cluster and 8-node IBM SP1. Best regards, Riccardo Valente -- Riccardo Valente e-mail: valente@hp720.sissa.it International School for Telephone: [+]39 40 3787428 Advanced Studies Home: [+]39 40 5700022 Via Beirut 2 [+]39 432 532252 I-34100 Trieste, Italy FAX: [+]39 40 3787528 FEGS Ltd Principal Software Engineer Oakington House Oakington Cambridge CB4 5AF England tel: +44(0)1223 237111 p.harriss@fegs.co.uk fax: +44(0)1223 234192 ------------------------------------------------------------------------------- From: Ian Jones My group use PVM in some parallelisation work we are doing. The application is Computational Fluid Dynamics. Machines are IBM SP1/SP2 and clustered workstations. Suggestions for improvements. Forgive my naive question, but I thought MPI was supposed to be the improvement for PVM, and the long term future. I have no suggestions for improvement, except to say that the most important issues are standardisation, robustness and speed. It is relativly easy to change message passing libraries, but I don't want lots of different versions, for different versions of PVM, MPI, PARMACS...... If it ain't broke, don't fix it. ******************************************************************* * * * Ian P Jones e-mail ian.p.jones@aea.orgn.uk * * CFDS phone +44 235 43 2464 * * B8.19 Harwell Laboratory fax +44 235 433 6671 / 3174 * * Oxon OX11 0RA * * UK * * * ******************************************************************* ------------------------------------------------------------------------------- From: aler@soleil.serma.cea.fr (Gregoire ALLAIRE - LETR) Yes, I use PVM. Applications : domain decomposition method in fluid dynamics. Machines : a network of workstations, or a combination of a CRAY C94 host) and a CRAY T3D (slaves). Gregoire Allaire CEA (French Atomic Energy Commission) N.B. Any opinions expressed here are solely mine, and not of my organization. ------------------------------------------------------------------------------- From: petters@lpr.e-technik.tu-muenchen.de (Stefan Petters) Dear Colleagues, as my english isn't the best, I apologize in advance for the various mistakes. We are working on an offline task-allocation-manager for a multiprozessor- realtime-system. The goal is to find an as cheap a possible allocation of the tasks on this system without any deadlines. The whole task-allocation- manager is separated in four parts: 1. grafical editor (for defining the system) 2. alalyzer (for checking all deadlines) 3. optimizer with simulated annealing 4. optimizer with genetic algorithms (GA) Since the number of possible sollutions is to big to compute complete, the given heuristics are used. The GA-optimizer is my part to implement. To compete with the simulated annealing in speed, I'm trying to spawn each Individual of the GA on one extra computer. GA is based on the evoluation process found by Darwin. Each Individual contains an task-allocation-table (TAT), exchanges with other Individuals genes (parts of that TAT), mutates slightly and selects the fitest for forming the next generation. I first tried sockets and PVM and have decided to use PVM. In the next week we are trying to move the editor aout of the programm in an extra task as it is bound to LINUX. The used computers are: 3 DEC alphas 3 LINUX (i486) 5 IBM 6000 (RS6K) 4 SUN4 2 MIPS I was amzed how easy the PVM-system was installed on all our computers. One thing I'm not happy about is that LINUX-machines need somtimes about one minute to follow the pvm_addhost call, but I fear this is a LINUX-problem and not yours. Sometimes the PVM-deamons don't follow an halt command, and loop abandoned on the host. I'm trying a workaround with a shell-script. An security system watching over lost hosts and lost tasks would be a fine thing. As conclusion I want to thank you for PVM as it has spared me much frustration with sockets. I hope I could help you. Best wishes Stefan Petters Email: petters@lpr.e-technik-tu-muenchen.de Phone: ++49-89/2105-3549 ------------------------------------------------------------------------------- From: laurent giraud Hello from France, I am not using PVM yet, but only amoug 2 hours a week, my appllication running under PVM is cellular automaton for hydrodynamics (Lattice Boltzmann method). I have 5 HP 712 (HPPA), 1 Silicon G (SGI5) and a Linux (Pentium 90 with Intel bug!!) under PVM. Suggestion:It would be wonderful if one day I can pick up a part of an array and directly send it to an another process!! Sorry for my very bad english... -- ******************************************************************************* * Laurent GIRAUD * * _/_/_/ _/_/_/ _/_/_/ _/_/_/ * * APPLICATIONS SCIENTIFIQUES _/ _/ _/ _/ _/ _/ * * DU CALCUL INTENSIF _/ _/ _/ _/ _/ * * C.N.R.S. _/_/_/_/ _/_/_/ _/ _/ * * Batiment 506 _/ _/ _/ _/ _/ * * Universite Paris Sud _/ _/ _/ _/ _/ _/ * * 91405 ORSAY cedex FRANCE _/ _/ _/_/_/ _/_/_/ _/_/_/ * * Tel : 69.82.41.97 * * * * e-mail: lgiraud@asci.fr * ******************************************************************************* ------------------------------------------------------------------------------- From: Werner Hett > Are you using PVM today? No, but maybe next week. > > What are your application(s)? - Parallel programming course (undergraduate level). - We have developed a library of C functions called PPC (Portable Parallel C) which is based in part on PVM. For further details see http://www.info.isbiel.ch/dpt/info/parallel/ > How many and what kind of machines have you used with PVM? About 30 Suns (Solaris 2), a few NeXTs, lots of PCs with Linux. > Any comments and suggestions for improvement are welcome. > Please email the information to pvm-survey@cs.utk.edu. Support of light weight processes or threads would have helped us a lot (of course this is not only a problem of PVM, we know!). We appreciate your work very much. Thank you for your effort! Best wishes, Werner Hett --------------------------------------------------------------------- Werner Hett, Ingenieurschule Biel, Postfach 1180, CH-2501 Biel-Bienne Tel +41-32-266 302 Fax +41-32-234 377 hett@info.isbiel.ch Switzerland --------------------------------------------------------------------- ------------------------------------------------------------------------------- From: Andrea OMODEO > Are you using PVM today? Yes > What are your application(s)? We are simulating power plant simulators on a network of workstations. Our work has been presented at the 1994 European PVM Users Group Meeting > How many and what kind of machines have you used with PVM? 1 SUN, 2 RS/6000, 2 DEC ALPHA We can use an IBM SP/1 (8 nodes) for testing. > Any comments and suggestions for improvement are welcome. We badly need a LAN-based version of PVM!!!. Andrea Omodeo & Alfredo Femminella ----------------------------------------------------------------------------- |Andrea Omodeo e-mail: omodeo@cra.enel.it | |ENEL - Centro Ricerca di Automatica andrea@omodeo.cra.enel.it | |v. Volta 1 - 20093 Cologno Monzese (MI) http://www.cra.enel.it/~andrea/ | ----------------------------------------------------------------------------- ------------------------------------------------------------------------------- From: (Niandong Fang)fang@ifi.unibas.ch 1. Are you using PVM today? We are still using PVM today but for student's course. 2. What are your application(s)? In the last year, we used it to develop some examples as - matrix multiplication, - warshall alg - game of life - stencil 3. How many and what kind of machines have you used with PVM? Workstation cluster (NeXT, SUN, VAX), SP2, etc Comments: We are changing our PVM program to MPI, as MPI seems ad hoc standard in the future for message passing. ------------------------------------------------------------------------------- From: Markus Baum Helo, to answer your questions: Yes i am using PVM today (actually i am using PVM nearly every day). I am running my own CFD code for direct numerical simulation of Navier Stokes equations. Actually i am using SP1, SP2 and RS6000 clusters using FDDI in configuration of 1 to 6 processors. Bye, bye Baum ------------------------------------------------------------------------------- From: auge@msmfs1.mw.tu-dresden.de (Andreas Auge) We are using PVM 3.3.4 today. We are developing CFD code (FEM, Domain Decomposition). We used not more than 7 machines in parallel. Our code runs on PC386/486 (LINUX), SUN SPARCstations (SunOS 4.1.2) HP 9000/735 (HP-UX 9.05), DEC ALPHA (OSF/1). We used these machines heterogen in parallel. Comments: unfotunately PVM is not available (public domain) for the German Transputer systems (Parsytec) based on T805 Transputers and PowerPC Chips. And, the commercially available products (PVM/PARIX - Genias software, Germany) do not work satisfactory and are only PVM 3.2 compatible. Best wishes, A. Auge ------------------------------------------------------------------------------- From: "Stephane.Vialle" >Are you using PVM today? Not today, but we have used it last year. >What are your application(s)? We have used PVM for student projects, they have discovered parallelism easily with PVM. We work on the ParCeL project: a PARallel CEllular Language for AI. Today it runs on MIMD architectures with a dedicated virtual machine. But we want install ParCeL on PVM and make benchmarks in order to compar with our virtual machine. >How many and what kind of machines have you used with PVM? We have used PVM on sparc stations. Stephane Vialle (vialle@ese-metz.fr) ------------------------------------------------------------------------------- From: Dietmar Lammers > > Are you using PVM today? yes (or at least tomorrow ...) > > What are your application(s)? I am implementing a distributed object oriented system (MOM), which uses PVM as it's base routing system. > > How many and what kind of machines have you used with PVM? SUN4 and SUN4SOL2. Not more, because my system uses cmu-CommonLisp, and in this state of the Implementation I've installed cmucl only under this both OSs. > > Any comments and suggestions for improvement are welcome. To avoid overhead with process grouping, I use a pseudo-broadcast to all running PVM-processes (calling pvm_tasks(0,...) first, then sending multicast to each host). This may be a useful function for the PVM system. I don't know how to get a unique TAG for my messages -- couldn't PVM provide some function for this purpose? ------------------------------------------------------------------------------- From: Jean-Luc Pons >Are you using PVM today? - Yes >What are your application(s)? - Molecular Modelisation; Distance Geometry... >How many and what kind of machines have you used with PVM? - 4 machines Hewlett Packard (3*735,715) , for intence - I'm going to add Silicon Graphics machines (4 Indigo) jlPons@cbs.univ-montp1.fr ------------------------------------------------------------------------------- From: dag.bjorge@dnmi.no (Dag Bjoerge) Are you using PVM today? Yes What are your application(s)? Limited Area weather forecast models. How many and what kind of machines have you used with PVM? Development on Cray YMP. Dag Bjoerge DNMI (Norwegian Meteorological Institute) ------------------------------------------------------------------------------- From: Nigel Starling > Are you using PVM today? > The IT Service at the University of Durham, UK is participating in research to schedule PVM tasks using CONDOR. Other research groups in the university also use PVM. > > What are your application(s)? > Two demonstration applications have been ported to use PVM in a CONDOR environment. 1. Non-Linear Least-Squares functional evaluation. 2. Image Rendering. > > How many and what kind of machines have you used with PVM? > HP-UX 9.0x on a pool of 1 x HP9000/750, 4 x HP9000/730 50 x HP9000/710, --- Nigel =============================================================================== Nigel Starling Tel: 0191-374-2872 Research Programmer Fax: 0191-374-3741 IT Service - University of Durham Science Site, South Road DURHAM DH1 3LE Mail:N.A.Starling@durham.ac.uk =============================================================================== ------------------------------------------------------------------------------- From: Fabien Coelho > Are you using PVM today? Yes, I am. I use PVM as a message passing library for a prototype HPF compiler. > What are your application(s)? Nice dataparallel applications. Seismic codes... > How many and what kind of machines have you used with PVM? Network of a few Sparc Stations, and some tests were run on a CM5. Fabien. -- Fabien COELHO __ http://www.cri.ensmp.fr/~coelho __ coelho@cri.ensmp.fr CRI, ENSMP, 35, rue Saint-Honoré, 77305 Fontainebleau cedex, France vox: 33.1.64.69.48.52, fax: 33.1.64.69.47.09, std: 33.1.64.69.47.08 ------------------------------------------------------------------------------- From: rcarbonell We are using PVM in seismic processing applications in particular: Kirchoff Pre-Stack depth migation. I have implemented the program with PVM calls and I am using 9 IBM RISC 6000, different models I works pretty good. ------------------------------------------------------------------------------- From: "Paul Gipson" > > Are you using PVM today? > YES > What are your application(s)? > CHEMISTRY, CFD, FEA + others > How many and what kind of machines have you used with PVM? > SGI multiprocessor servers > Any comments and suggestions for improvement are welcome. The shared-memory implementation is extremely unreliable and essentially not usable. -- ........................................................................... . . . . Paul Gipson . Opinions are mine, . . Sales Support Mgr, High Performance Systems. not my paymaster's. . . Email: paul@reading.sgi.com . . . Tel: +44-1734-257618 (direct) . Cracking Toast, Gromit ! . . Voicemail: +44-1734-257501 (7618#) . . . Fax: +44-1734-257505 . . ........................................................................... ------------------------------------------------------------------------------- From: zfb@ibmpar.elis.rug.ac.be (Zhang Fu-Bo) > > Are you using PVM today? Yes. > > What are your application(s)? We developed an environment for automatical generation of PVM code from a serial application. We use Gauss-Jordan elimination and Matrix Multiplication to test PVM code. > > How many and what kind of machines have you used with PVM? > Normally, 4-12 machines are used. There are SUN Sparcs, IBM AIX, HP computers. > Any comments and suggestions for improvement are welcome. > Please email the information to pvm-survey@cs.utk.edu. > Yes, we find if PVM supports shared memory address space, it is much easy to write users' applications. Best regards, fubo ZHANG ------------------------------------------------------------------------------- From: Josep Maria Garrell i Guiu > > Are you using PVM today? Yes. > > What are your application(s)? Paral.lel simulated annealing. > > How many and what kind of machines have you used with PVM? LINUX, HPPA, SUN4. ------------------------------------------------------------------------------- ------------------------------------------------------------------------------- From: sven.graupner@informatik.tu-chemnitz.de (Sven Graupner) Dear PVM-Team, > Are you using PVM today? Yes, we've just started ''playing with it'' > What are your application(s)? One of my students investigates performance benefits of a ''parallel make'' application in a network of workstations, which runs 'make -n' and distributes the generated command lines via PVM and carries them out remoteley. \\ Has any comparable work already been done somewhere? Is there a parallel make application? > How many and what kind of machines have you used with PVM? We have a Pool of 25 Sun/ELC,IPC workstations. > Any comments and suggestions for improvement are welcome. We have just started working with PVM and it really seems easy to use. Wishes for improvements will shurely rise while working with PVM. If I have ideas, I'll let You know. Good luck, Sven -- ------------------------------------------------------------------------ Name: Dipl.Inf. Sven Graupner Institute: Technische Universitaet Chemnitz Fakultaet fuer Informatik, Lehrstuhl Betriebssysteme Address: Strasse der Nationen 62, 09111 Chemnitz, Germany PF: 964, 09009 Chemnitz Phone: (+49)-371-531-1390 Fax: (+49)-371-531-1530 e-mail: sgr@informatik.tu-chemnitz.de ------------------------------------------------------------------------ ------------------------------------------------------------------------------- From: Bill Pearson Hello, just a quick reply to the survey. Yes I am using PVM today. I am using it to switch between one of three batch programs in an interactive flow solver. With a view to expanding it into a parrallel machine at a later date The machines I am using so far are sparc2 and sparc10 (running solaris 2.3) and dec 5000 machines. Thanks for the code. Bill. ------------------------------------------------------------------------------- From: umberto@morino.ing.uniroma1.it (Umberto Iemma) 1) Yes, I am currently using pvm version 3.2. (I am going to install the new version, together with the first release of Xpvm). 2) My activity deals with CFD application using Boundary Integral Equations approaches (very high parallel efficiency) in the transonic range. 3) I use: RISC 6000 mods. 320/360/375/3AT SGI INDY/INDIGO2 IBM SP2 (not yet implemented) Bye Dr. Umberto Iemma Ph.D. in Aerospace Engineering ------------------------------------------------------------------------------- From: kowallik@ifh.de (Rainer Kowallik) > > Are you using PVM today? Not today and propably not this week, but I have used it in the past and will need it in the future. However, my plannings are going in the MPI Direction. > > What are your application(s)? We have an IBM-SP2 (10 nodes, HPS) and an SGI-challenge (14 CPUs, SHM) which are used for calculations, data aquisition, data analysis and simulation in High energy physics. We have successfully converted a simulation program for muon showers and are working on some kind of half automated process which aids physicists in converting existing FORTRAN code into a parallel form. Plans are to use parallel computers for Data aquisition. Up to now we are investigating different models of parallel programming and theier use for our applications. For testing purposes we have develloped a PVM-driver for MPICH and a SHM-driver for MPICH with the aim to test, if we would be able to use MPI for special real time parallel processors. > > How many and what kind of machines have you used with PVM? An IBM-SP2 with 10 nodes and HPS, over IP emulation and standard Ethernet. An SGI-challenge with 14 processors and shared memory. > > Any comments and suggestions for improvement are welcome. I dislike the way PVM forces a user to make links in ~/pvm/arch to the executable. Although I can see that you need such a mechanism for a heterogenous environment it would be nice, if PVM would just take the path to the actual executable if all architectures are the same. This would ease the task for a computing center to provide simple users with parallel applications. The Shared memory implementation of PVM still needs some attention. The situation has improved but is far away from being perfect. I would not give any parallel application based on PVM-SGIMP to our normal users. The risc would be too high, that the machine resources will be blocked completely. Apart from my critics above PVM is really a very nice and efficient tool. Thank you for the work you have done. Rainer Kowallik, Postdoc at DESY-IfH Zeuthen. kowallik@ifh.de ------------------------------------------------------------------------------- From: frolkovic@fmph.uniba.sk (Peter Frolkovic) Hi! I am "maintainer" of PVM at our Faculty of Mathematics and Physics (Comenius University Bratislava, SLOVAKIA) and I got (forwarded from sk2eu.eunet.sk) your questions about using PVM. Are you using PVM today? ======================== Yes. Especially one my graduate student is now working intensively with it. What are your application(s)? ============================= They are more "pedagogical". Student is testing different Gauss-Seidel methods for problem arising from 1D parabolic PDE, I plan (if ...) to work with the same for 3D problem (with PLTMG in FORTRAN from Netlib). How many and what kind of machines have you used with PVM? ========================================================== Once I tried 8 different DECstations with Ultrix and 7 SUN SparcStations with Solaris. Any comments and suggestions ... ================================ PVM (and especially XPVM) is working well for us. If some things will be decided about my future in some manner, I plan to devote more time to PVM and I am sure it will be used by more and more people at our faculty. Continue with your nice and useful work! With the best wishes Peter Frolkovic Department of Numerical Analysis Comenius University Bratislava SLOVAKIA frolkovic@fmph.uniba.sk ------------------------------------------------------------------------------- From: I'am using PVM for developing inter process commnucation in actor Langage named MeringIV (with load balancing module). actually a prototype of MeringIV runs on SUN4 machines. Carle Patrice DES/SIA O.N.E.R.A 29 Av. de la division Leclerc B.P 72 92320 Chatillon CEDEX addresse elec: carle@onera.fr tel : 46-73-43-72 fax : 46-73-41-49 ------------------------------------------------------------------------------- From: Malte Gross Thank you for your mail message asking about the use of PVM. Yes, I use PVM for about 2 months now. I wrote a code performing a 3-dimensional FFT on a cubic grid. The cube is divided into 2-D slices, each one residing on a single processor. This code is part of a larger project pointing toward the solution of Maxwells equation for a given charge distribution. Up to the now, I mainly used an emulator on a SUN4, but I'm turning over to a CRAY T3D with 128 processor nodes for the production runs. I hope that this information will be useful for you. M. Gross ------------------------------------------------------------------------------- From: frelle@math-tech.dk (Lars Frellesen) Are you using PVM today? yes What are your application(s)? A parallel direct linear equation solver An oil reservoir simulator. How many and what kind of machines have you used with PVM? 1 Sparc 10 SunOs 4.1.3 machine Lars Frellesen __________________________________________________________________________ Math-Tech Aps E-mail : frelle@math-tech.dk Kildeskovsvej 67 2820 Gentofte Tlf : +45 31 63 69 88 DK - Denmark Fax : +45 31 68 31 95 __________________________________________________________________________ ------------------------------------------------------------------------------- From: Kees Kassels PVM Project Team, - We are using PVM for our parallel application. We are solving the Navier-Stokes equations ( fluid dynamics) using domain decomposition. These domains are distributed to different machines using PVM. - We have used 20 machines all HP725 or HP735 -- Kees Kassels Fac. of Technical Mathematics and Informatics Delft University of Technology Mekelweg 4 P.O. Box 5031 2600 GA Delft The Netherlands Tel. (31) 15 - 783639 Fax. (31) 15 - 787209 E-mail: c.g.m.kassels@math.tudelft.nl ------------------------------------------------------------------------------- From: mem1001@cus.cam.ac.uk (M.E. Mura) > Are you using PVM today? As a basis for mpi through MPICH. > What are your application(s)? Quantum Chemical codes. > How many and what kind of machines have you used with PVM? Cray T3D (2) SGI workstations indigos (4) -- Dr Michael Mura, PD Research Fellow, Theoretical Department,Univ. Chem. Labs., Lensfield Rd, CB2 1EW, Cambridge, UK. memura@theor.ch.cam.ac.uk, mem1001@cus.cam.ac.uk, ph +44-223-336423 (fax 336362) ------------------------------------------------------------------------------- From: Maarten Hagen Here the answers to the posed questions: 1) No, I am not using PVM today, 2) The applications are: - computations of phase diagrams - computation of viscous flow. 3) We have used PVM on 13 machines, all Silicon Graphics: About 7 RS4000, and 6 RS600. -- Groeten Maarten 020-6081229 (AMOLF) 020-6231759 (THUIS) ------------------------------------------------------------------------------- From: berthold@maral.kfunigraz.ac.at (Guenther H. Berthold) It was dark and stormy on Jan 6, 7:04pm, when pvm-survey@cs.utk.edu wrote: > Subject: PVM survey your message> Dear Colleague, your message> In order to improve PVM we are trying to understand how your message> people are using the system. Your feedback is important. your message> Could you take a moment to answer three questions. your message> your message> Are you using PVM today? Yes your message> your message> What are your application(s)? Application for research in the following areas: theoretical chemistry (GAMESS) high-energy physics solid state physics mathematics your message> your message> How many and what kind of machines have you used with PVM? about 13 your message> your message> Any comments and suggestions for improvement are welcome. Since we are administrating PVM by the central computing center, we would like to see more administration tools for load balancing and log file handling. your message> Please email the information to pvm-survey@cs.utk.edu. your message> your message> Your help is much appreciated. your message> your message> Best wishes, your message> The PVM Project Team your message> your message> ps. Previous survey results are available through World Wide Web at your message> http://www.epm.ornl.gov/pvm/ your message> your message> pps. We apologize if you received multiple copies of this message. your message> Our first attempt at sending the message uncovered a bug in our mailer. >-- End of excerpt from pvm-survey@cs.utk.edu -- ***************************************************************************** * Guenther H. Berthold * * EDV-Zentrum email: guenther.berthold@kfunigraz.ac.at * * Karl-Franzens-Universitaet Graz X.400: berthold@edvz.uni-graz.ada.at * * Universitaetsstrasse 27 phone: +43 316 380 2241 * * A-8010 Graz fax: +43 316 326037 * * Austria - Europe * ***************************************************************************** ------------------------------------------------------------------------------- From: botorog@zeus.informatik.rwth-aachen.de (Georg Botorog) Hello, Although I haven't yet used PVM, I might do it sometimes in the (not very near) future. I plan to implement a parallel language with algorithmic skeletons. Therefore, the idea of of building upon PVM is not central to my pursuit. I have thought of the PVM (or maybe MPI, but PVM seems to me more mature at the time) interface as a means to implement my language on different architectures, mainly a Parsytec transputer network and a Sparc workstation cluster. I am aware that the use of a standard library will lead to a slow-down of my programs, and since I have to get the best possible speedups, the standard library is, as I already have said, not central to my project. Nevertheless, depending on how things will evolve, I will keep you informed on any progress. To summarize: * I am not using PVM today, but might do so in the future * I am planning to implement a parallel language with algorithmic skeletons * The available hardware consists at the time of a transputer network (T800's from Parsytec) and a bunch of Sun Sparc workstations. Furthermore, some other machines, such as an Intel Paragon, might be available. I truly doubt that my answer will be of importance to you, but if anything changes, I will contact you. Best regards, Horatiu Botorog ------------------------------------------------------------------------------- From: Yves Epelboin > Are you using PVM today? yes > > What are your application(s)? Simulation of white beam synchrotron topographs. This is an application of the Dynamical Theory of X-Rays > > How many and what kind of machines have you used with PVM? 3 only: 2 IBM + 1 SGI ------------------------------------------------------------- : Yves EPELBOIN : +-----------------------------------------------------------+ : Lab. Mineralogie-Cristallographie, UA 09 CNRS, : : Case 115, 4 place Jussieu : : F-75252 PARIS CEDEX 05 : : phone: 33 1 44 27 52 11 : : Fax: 33 1 44 27 37 85 : : E-Mail: epelboin@lmcp.jussieu.fr : : WWW : http://www.lmcp.jussieu.fr/~epelboin : +-----------------------------------------------------------+ : UNIVERSITE PIERRE ET MARIE CURIE PARIS FRANCE : +-----------------------------------------------------------+ ------------------------------------------------------------------------------- From: "Derek Nitch" In reply to your request for information : > Are you using PVM today? Yes. > What are your application(s)? Solution of Maxwells equations using method of moments. Analysis of electromagnetic structures. > How many and what kind of machines have you used with PVM? 3 Sparc 10's 1 RS6000 > Any comments and suggestions for improvement are welcome. > Please email the information to pvm-survey@cs.utk.edu. Development and debugging of our code was done under OS/2 on a single 486-66. No problems were encountered during this stage of the development. However we could not get a network of OS/2 machines to work. PVM failed to connect to / locate the other processors. Unfortunately, I do not have the OS/2 TCPIP developers kit and I am therefore unable to find out what the problem is. All the best Derek Nitch. ********************************************************************** * Derek Nitch Dept. of Electrical Engineering * * University of the Witwatersrand Computational Electromagnetics * * Johannesburg nitch@odie.ee.wits.ac.za * * South Africa * ********************************************************************** ------------------------------------------------------------------------------- From: ilpide@CS.UTK.EDU (Ilpide) Dear sir, We've been using PVM for a few months now. The target program is a finite element one first developped for a shared memory multiprocessor and adapted for PVM using a SPMD model. The PVM program runs on a SGI cluster (we've used up to 4 of them) and has been compiled on a single LINUX based PC. We will, in the near future use this program on a small PC network (probably with LINUX). Programs are written in fortran and good speedups were obtained. If you ever need more information please just e-mail me. I'll be glad to give you more details. Yours sincerely, E. ILPIDE #--------------------------------------------------------------------------# EMMANUEL ILPIDE | Tel : (+34) (3) 401 64 85 Centro Internacional de Metodos | Fax : (+34) (3) 401 65 17 Numericos en Ingenieria (CIMNE) | E-mail : ilpide@agripa.upc.es Barcelona - SPAIN | vilpide@etseccpb.upc.es ***** IRC : Manolo (Not the one you knew) #--------------------------------------------------------------------------# ------------------------------------------------------------------------------- From: > Are you using PVM today? No - I have changed jobs > What are your application(s)? Currently hydrological (and graphics) Previously parallel processing in general. > How many and what kind of machines have you used with PVM? HP700s (with HPUX 9.01) and Suns (with OS 4.1.x) Mixed or a single type. I only used a small number of machines but I could have used up to 16 Suns and 60+ HPs. Ann Petrie ----------------------------------------------------------------------- Ann Petrie Ann.Petrie@ncl.ac.uk Water Resource Systems Reseach Unit Room: 302, Cassie Building Department of Civil Engineering Phone: +44 - 91 - 222 - 5614 University of Newcastle upon Tyne Fax: +44 - 91 - 222 - 6669 Newcastle upon Tyne, NE1 7RU Telex: 53654 UNINEW G ----------------------------------------------------------------------- ------------------------------------------------------------------------------- From: Roger Lindell Are you using PVM today? Not at the moment, but we are planning to try it. What are your application(s)? Speech recognition using amongst other things artificial neural nets. How many and what kind of machines have you used with PVM? Planning to use it on about 18 HP 9000/7XX computers. ------------------------------------------------------------------------------- From: Pascal Amand Hi, I am using PVM3.3.2 in the field of seismic ray tracing in order to perform tomography. The parallel architecture is master/slave, using a farming strategy and automatic load-balancing. I experimented PVM both on networked SUN4 workstations and CM5. regards Pascal -- Pascal Amand CNRS, Institut de Geodynamique (Equipe RUaDE) 250 Avenue Albert Einstein, Sophia-Antipolis, 06565 Valbonne, FRANCE Tel : 93.95.42.54 - Fax : 93 65 27 17 - Email : amand@faille.unice.fr ------------------------------------------------------------------------------- From: Hayat L 1. yes, we are using PVM. 2. We are developing a library of parallel image processing algorithms. 3. We have used up to 24 SUN4, 1 Dec ALPHA, 1 ButterFly, and 2 SiliconGraphics. ------------------------------------------------------------------------------- From: Dear Colleague, We compiled PVM for two SUN Sparcstations (type 5 and 20) and HP (apollo 700). We plan to compile it also for SGI indy. It works O.K., once the PVM-deamons are started, but halting gives problems. Often if PVM is halted on the main machine, the deamons on the other machines don't stop. I then have to kill the processes manually and remove the files pvmd.xxx (Both XPVM and the PVM-console have this) This problem occurs on all machines. We plan to use PVM in the MIAMI project (http://kunpu7.psych.kun.nl/miami/), because it looks promising for communication between agents on different hosts. Greetings, Jan Nijtmans ------------------------------------------------------------------------------- From: diethelm@informatik.uni-hildesheim.de (Kai Diethelm Mitarbeiter Prof. Alten) Dear Colleagues, I would like to answer your questions in the PVM survey as follows: We are running PVM on a network of twelve Sun SPARCstations (different types: SPARCstations 1, 1+, 2, iPC, 10). However, we have only just started this, and we are still more or less learning how to use it for ourselves. We do not have any 'serious' applications yet, just some experiments to se how parallelism works. If we are successful in teaching PVM to ourselves, then we plan to teach it to graduate students in the context of parallel algorithms in numerical analysis, but this is going to be somewhen in the future (no specific date yet). Yours, Kai Diethelm. --------------------------------------------------------- Dr. Kai Diethelm Institut fuer Mathematik Universitaet Hildesheim Marienburger Platz 22 D-31141 Hildesheim Germany e-mail: diethelm@informatik.uni-hildesheim.de ------------------------------------------------------------------------------- From: > Are you using PVM today? Yes, I use pvm every day quite heavily. >What are your application(s)? I am curently building a SPMD parallel version of our soft. RADIOSS. RADIOSS is a Finite Element package for mechanical enginieering, used mainly to simulate car crash (but also train, planes...). The average CPU for a full car simulation is in the 10 hours range on a CRAY YMP and can go up to about 150 hours. Hence we (and our customers) are very interested in parallel solutions and we build already a shared memory and a data parallel ve parallel version. The SPMD version should become industrial end of february. We already achieve speedups of 14.9 on 16 prcs on SP2 for academic tests. >How many and what kind of machines have you used with PVM? Right now we have worked only on IBM platforms (SP1 SP2) but we will try other platforms within the next quarter and plane to replace the shared mem. versio version with the SPMD one. Regards, Dimitri Nicolopoulos Manager of the parallel applications MECALOG ------------------------------------------------------------------------------- From: ruud@wop.wtb.tue.nl (Ruud Eggels) I am not using PVM today, but I have used it, however, with succes for a program that computes reduced reaction schemes, which can be used for modeling laminar and turbulent combustion processes. Uptil now I have used PVM up to 8 Silicon Graphics machines (Indy, Indigo). ------------------------------------------------------------------------------- From: j.watson1@physics.oxford.ac.uk (John Watson) > > Are you using PVM today? Not in the last month. > > What are your application(s)? Large floating point calculations - numerical solutions of differential equations. > > How many and what kind of machines have you used with PVM? > Suns - Sparc 5's, Sparc 10's and Sparc Classics Dec Alphas. -- =============================================================== Name: John Watson Telephone: Lab:(0865)272260 Home:(0865)515425 e-mail: j.watson1@physics.ox.ac.uk "It wasn't blood in general he couldn't stand the sight of, it was just his blood in particular that was so upsetting." -(Terry Pratchett, Sourcery) =============================================================== ------------------------------------------------------------------------------- From: Dear Sir Yes I am using PVM today Simulation of Free Surface Flows 8 Sparc ones J. A. Cuminato ------------------------------------------------------------------------------- From: Stefan Lang To whom it may concern, 1. We are using PVM in the implementation phase of our new parallel algorithms. Furthermore when comparing speedups on different parallel architectures we also consider a workstation cluster communicating with PVM. 2. Our applications are parallel numerical algorithms for the solution of partial differential equations.The features provided are unstructured grids, local mesh refinement and adaptivity in the solution cycle. One parallel aspect of interest is the redistriution of the mesh on the processors gaining an equal load and low communication needs. In this field we develop fast, efficient solvers for large sparse matrices, especially we use the multigrid approach. 3. Up to this time we ran PVM on clusters of SUNs, SGIs and PCs. I also just used the CRAY implementation of PVM on a CRAY T3D with 32 PEs. Hope to help you. Stefan ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ ~ ~ ~ Stefan Lang, Numerik fuer Hoechstleistungsrechner ~ ~ Institut fuer Computeranwendungen III ~ ~ Pfaffenwaldring 27 IIIII CCCC A 333 ~ ~ 70569 Stuttgart I C A A 3 ~ ~ I C A A 333 ~ ~ phone: 0711 / 685-7003 I C AAAAA 3 ~ ~ email: stefan@ica3.uni-stuttgart.de IIIII CCCC A A 333 ~ ~ ~ ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ ------------------------------------------------------------------------------- From: Alain Dupuy Dear PVM people, I am in charge of high performance computing applications in one of Elf (the french Petroleum company) research lab in London. We have developped a parallel reservoir simulator and we do use PVM to run it. It is useful to us in 2 ways, first to run on our cluster of up to 8 HP workstations (720, 750 and 735) and then because it is supported by all vendors so we can run tests on their machine with little modification to the code. (Cray, Convex and IBM namely). Regarding future directions of PVM, we are not using xpvm to monitor our parallel programs. I have noticed a few shortcomings, first it is very slow when the density of communications is high. Then I think it would be great to display the source code by clicking on the time bar for instance as it is done for the messages. Best Regards, Alain Dupuy ------------------------------------------------------------------------------- From: Kris Demuynck > What are your application(s)? We are developing a software development system for Virtual Reality applications. This system is based on processes and communication between them. More info can be found on the WWW-page at URL: http://www.ruca.ua.ac.be/Memex/VEplatform/ > How many and what kind of machines have you used with PVM? We are using the following computers: - HP 9000/735 workstation with HP/UX 9.01 - HP 9000/712 workstation with HP/UX 9.05 - SUN Sparc2 with SunOS 4.1.3 - PC 486-40MHz with Slackware Linux 1.2.0 All computers are linked in an ethernet LAN. > Any comments and suggestions for improvement are welcome. PVM is truly a very good system for distributed processing. We have been able to adjust our software (which ran on one computer) to PVM within a half day! Also, the program compiles on every machine we use without any error. This means that PVM accomplishes some of the highest demands, being PORTABLE ! However we also experience that the fact that PVM is build on existing SLOW network protocols, the latency is very high. This is also because we use lots of small messages (200-500 messages/second with 50-100 bytes each). An improvement for small messages would be a great breakthrough. We will continue to evaluate PVM versions in the future and use it in our software development. Kris. ------------------------------------------------------------------------------- From: Dear collegue, I appreciated your mail concerning the PVM survey. Although we have not yet installed PVM, I guess our answer could be useful to you. >>Are you using PVM today? >> No, we have not yet installed it, but we plan to do this within this year. >>What are your application(s)? >> We intend to use PVM to speed up a finite difference code for acoustic/elastic seismic modeling, developped by my R&D group. If this will effectively improve the computation, we intend to apply the same technique to a program for finite difference imaging of seismic data in complex media. >>How many and what kind of machines have you used with PVM? >> We wish to install PVM on some (2-4) SPARC 10 SUN workstations, as well as on a forthcoming SPARC server with 4 CPU and 512MB of shared memory. I thank you for your efforts in improving PVM, and I send you my best wishes for 1995. Yours truly Vittorio De Tomasi AGIP R&D ------------------------------------------------------------------------------- From: Till Bubeck Are you using PVM today? Yes. What are your application(s)? We're developing a system for distributed computing following the fork/join paradigm on shared-memory multiprocessors. A more complete description of this system called "Distributed Thread System" (DTS) can be found at http://ti-www.informatik.uni-tuebingen.de/~bubeck/doc/HPCN.ps.gz. How many and what kind of machines have you used with PVM? Up to 40 Sun Sparc's running SunOS 4.1.3 and Solaris 2.3. Any comments and suggestions for improvement are welcome. PVM should be made thread-safe. This means, that the message buffer management routines should handle multiple buffers at once. The distinction could be made by specifying the buffer as an argument to all pvm_pack*() and pvm_unpack*() functions. ,,, (o o) -----------oOO--(_)--OOo------------------------------------------- | Tilmann Bubeck | _||__ | | Universitaet Tuebingen | / . | <- Berlin | | Technische Informatik | | _/ | | Sand 13, 72076 Tuebingen | |_ \ | | Germany | /*__/ <- University of T"ubingen | |------------------------------------------------------------------ | Tel: +49 7071 295865 Fax: +49 7071 610399 | | Email: bubeck@peanuts.informatik.uni-tuebingen.de | | WWW/Mosaic: http://www-ti.informatik.uni-tuebingen.de/~bubeck | ------------------------------------------------------------------- ------------------------------------------------------------------------------- From: > Are you using PVM today? Yes. > What are your application(s)? Nothing concrete -- I've used it to write a package for distributing data-parallel operations around a network of parallel and serial processors. > How many and what kind of machines have you used with PVM? Stardent Titan, SUN4, SGI Indigo, Meiko CS-2, Maspar MP-1 __ Duncan Batey _____________________________ School of Mathematics, __ __ Postgraduate Research Student ______ University of Bath, England. __ __"I'm suffering from mood poisoning; it must be something I hate..."__ ------------------------------------------------------------------------------- From: Mathias Duemmler > > Are you using PVM today? > Yes, I'm using it for educational purposes. > What are your application(s)? > I'm supposed to write a Mandelbrot-generator. > How many and what kind of machines have you used with PVM? > I'm using a workstation-cluster with 10 DEC 3000 workstations. ------------------------------------------------------------------------------- From: In reply to your survey, yes I will be using PVM today. I am working on a trivial cellular automata at the moment but the main part of my project is to build a system to run on top of PVM to provide a framework into which many applications can be slotted just by changing small sections of code. I am using SUN SPARCstations provided by the university and the university computing department on two seperate networks, and there are about 50 machines in my hostfile I think. I hope this has been of some help to you and good luck for the future. Barry Scott ''' (o o) ---------------oOO--(_)--OOo---------------- ==========================================================================> BARRY SCOTT || u01bps@aberdeen.ac.uk OR COMPUTING SCIENCE DEPARTMENT || bps@csd.abdn.ac.uk KINGS COLLEGE || UNIVERSITY OF ABERDEEN || Latest excellent football result: OLD ABERDEEN || AIRDRIE 3 Vs 2 DUNDEE SCOTLAND, UK || B&Q Cup Final 6 Nov 1994 ==========================================================================> ------------------------------------------------------------------------------- From: rch@aix.fsv.cvut.cz (Rostislav Chudoba) Dear PVM team, We have started to use the PVM only recently on a cluster composed of three IBM RS/6000. We use also the HeNCE interface. Our applications area is the development of finite element algorithms for the analysis and optimization of structures. In the near future we will have the possibility to work on the SP2 equiped with 9 processors. Best regards and happy New Year ------------------------------------ Prof. Zdenek Bittnar, et. al Department of Structural Mechanics Faculty of Civil Engineering Czech Technical University in Prague ------------------------------------------------------------------------------- From: "Mr." Ian Masters Reply to your PVM survey mailing. >Are you using PVM today? Yes, now up to version 3.3.6. The system is administered by someone else for the department. >What are your application(s)? Finite Elements. There are about six people actively working on parallel processing here in the Department of Civil Engineering, Swansea, UK They are using PVM, MPI, and Unix for message passing. Personally I am using PVM to solve Heat Transfer/ solidification/melting problems. >How many and what kind of machines have you used with PVM? Originally we were using our SUN4 network over ethernet, but have recently installed a SGI iris network of four machines and a SGI challenge ( four processors, shared memory) using ATM connections. Hope that helps, Ian --------------------------------------- Ian Masters (cgmasters.pg@swan.ac.uk) Dept. of Civil Engineering, University of Wales, Swansea Swansea, UK --------------------------------------- ------------------------------------------------------------------------------- From: Till Bubeck ********************************************************************** * This is a correction of my previous mail, which had an illegal WWW * * address in it. Please ignore my previous mail. Thanks. * ********************************************************************** Are you using PVM today? Yes. What are your application(s)? We're developing a system for distributed computing following the fork/join paradigm on shared-memory multiprocessors. A more complete description of this system called "Distributed Thread System" (DTS) can be found at http://www-ti.informatik.uni-tuebingen.de/~bubeck/doc/HPCN.ps.gz How many and what kind of machines have you used with PVM? Up to 40 Sun Sparc's running SunOS 4.1.3 and Solaris 2.3. Any comments and suggestions for improvement are welcome. PVM should be made thread-safe. This means, that the message buffer management routines should handle multiple buffers at once. The distinction could be made by specifying the buffer as an argument to all pvm_pack*() and pvm_unpack*() functions. ,,, (o o) -----------oOO--(_)--OOo------------------------------------------- | Tilmann Bubeck | _||__ | | Universitaet Tuebingen | / . | <- Berlin | | Technische Informatik | | _/ | | Sand 13, 72076 Tuebingen | |_ \ | | Germany | /*__/ <- University of T"ubingen | |------------------------------------------------------------------ | Tel: +49 7071 295865 Fax: +49 7071 610399 | | Email: bubeck@peanuts.informatik.uni-tuebingen.de | | WWW/Mosaic: http://www-ti.informatik.uni-tuebingen.de/~bubeck | ------------------------------------------------------------------- ------------------------------------------------------------------------------- From: Xiaoming Zhang > > Are you using PVM today? Yes. > > What are your application(s)? Computational Fluid Dynamics. > > How many and what kind of machines have you used with PVM? 20 Sun Sparc-1s. ------------------------------------------------------------------------------- From: tvrdik@sun.felk.cvut.cz (Ing. Tvrdik Pavel CSc k336) > Are you using PVM today? Yes. > What are your application(s)? Semester projects = practical programming parallel algorithms using PVM3 - for the course Parallel Systems and Computations which is included into the curricula of Department of Computer Science at Czech Technical University, Prague. About 60 undergraduate students, grouped into 30 2-student teams. Size of the project is usually several 10KB of source code in PVM3 and C. > > How many and what kind of machines have you used with PVM? About 8 SUN4s. > Any comments and suggestions for improvement are welcome. UNix and/or PVM often crashes when all workstations are running PVM simultaneously, which leads often to frustration of the users and unkindness of the system administrator. In particular, the system get sometimes halted so that the keyboard is disconnected and the machine has to be physically switched off to be rebooted. Regards, Pavel Pavel Tvrdik Department of Comp.Sci.& Engr. Faculty of Electrical Engineering Czech Technical University Karlovo nam. 13, 12135 Praha 2 Czech Republic tel: +42 2 24-35-73-11 (office), +42 2 29-34-85 (secretary) fax: +42 2 29 80 98 email: tvrdik@sun.felk.cvut.cz ------------------------------------------------------------------------------- From: "Xavier Defago" > Are you using PVM today? Yes. > What are your application(s)? The runtime environment for a distributed pascal on the Cray T3D. The functionnalities of the T3D are simulated on a network of DEC Alpha workstations. > How many and what kind of machines have you used with PVM? 30 DEC Alpha AXP workstations running OSF/1 2.0 Cray T3D using an 8 PEs partition > Any comments and suggestions for improvement are welcome. > Please email the information to pvm-survey@cs.utk.edu. I happen on some occasions to crash the kernel while using PVM/XPVM intensively (i.e. about 300 PVM tasks scattered among 30 DEC stations, in trace mode.) I haven't been able to recreate the situation on purpose, though. PVM seems to be a very stable devellopment environment. Suggestions: ------------ - The use of an buffer ADT in order to avoid side effects with a global sbuf and global rbuf. (it proves to be a real annoyance and error prone with signals and threads). - A support for threads in a PVM task. - A support for PVM in the programming language (a preprocessor or the like) which would, for instance, allow to send complex structures without requiring the pack/unpack sequence. - It might already exist but a true Object-Oriented API to PVM could prove useful in the mid term (with objects duplication, object migration facilities, ...), or some OO fusion of PVM and the ISIS library develloped by Ken Birman at Cornell for distributed computing. - If a PVM micro kernel were implemented and the current PVM API one of many software layers running on top of it, the previous suggestion could be implemented as one of these software layers, and many more API could be tailored for different purposes. Happy new Year ! ------------------------------------------------------------------------------ Xavier DEFAGO e-mail: defago@di.epfl.ch EPFL-Switzerland www : http://diwww.epfl.ch/~defago/ ------------------------------------------------------------------------------- From: Helmut Heller Are you using PVM today? YES What are your application(s)? EGO, a molecular dynamics program How many and what kind of machines have you used with PVM? Workstation clusters (SUNs, NeXTs, SGIs, HPs, ALPHAs), CRAY T3D, IBM SP1, Power Challenge, PARSYTEC powerExplorer -- Servus, Helmut (DH0MAD) ______________NeXT-mail welcome_________________ FAX: +49-89-2394-4607 "Knowledge must be gathered and cannot be given" heller@nirvana.imo.physik.uni-muenchen.de ZEN, one of BLAKES7 Phone: +49-89-2394-4565, -4562---------------------------------------------- Helmut Heller, Ludwig Maximilians University Institute for Medical Optics, Theoretical Biophysics Group Theresienstr. 37, 80333 Munich, GERMANY ------------------------------------------------------------------------------- From: crubille@ceremab.u-bordeaux.fr (Paul Crubille) > What are your application(s)? Actually educational purpose only. > Are you using PVM today? Yes or rather our students are. > How many and what kind of machines have you used with PVM? About % Sun. Thanks for your work. P. Crubille ------------------------------------------------------------------------------- From: -> Are you using PVM today? Yes, I am actually using PVM vers. 3.3.4 and XPVM 1.0.3. -> What are your applications? I use PVM on simulations of Time Dependent Differential Equations and Cellular Automatas applications. -> How many and what kind of machines have you used with PVM? We run PVM on SUN machines, HP 7300, DEC AXP 2000-300, DEC AXP 210 A500MP, Convex 210 and, PC-386 running Linux. Best regards. -------------------------------------------------------------------------- Juan Carlos Fabero Jimenez Dpto. Informatica y Automatica Facultad de Ciencias Fisicas Universidad Complutense de Madrid (Spain) Telf. +34-1-3944384 E-Mail: ------------------------------------------------------------------------------- From: We are interested in porting PVM to a multicomputer machine based on Transputers from PARSYS. ------------------------------------------------------------------------------- From: Junxian Liu > Are you using PVM today? Yes. > What are your application(s)? We are writing a HPF to F90 mapper, which supports PVM as one of the message passing libraries. > How many and what kind of machines have you used with PVM? Currently, we run PVM on networked Sun workstations. Dr J Liu NA Software Ltd Liverpool L1 9DW U.K. ------------------------------------------------------------------------------- From: Tom Archibald Are you using PVM today? Yes What are your application(s)? Determining operating policies for reservoir networks Managing fuel contract portfolios How many and what kind of machines have you used with PVM? Up to 16 Sun workstations (SLC, ELC, Sparc2, Sparc5) ------------------------------------------------------------------------------- From: "C.E.Dean" 1. I have programs waiting to be modified to run under PVM, but pressure of other general sysadmin work has pushed the PVM work down in priority. 2. Four programs so far, related to Synchrotron Radiation research, i. Small Angle X-ray Simulation. ii. Calculation of the mass projection of a model structure. iii. Diffraction pattern simulation. iv. Background subtraction and Peak fitting of experimental diffraction patterns. "Computing with Parallel Virtual MAchines", C.E. Dean, R.C. Denny, P.C. Stephenson, G.J. Milne and E. Pantos, J. de Phys. III, Colloque C9, 445-448, November 1994. 3. I have been using a group of Silicon Graphics machines, Indy, Indigo2 and 2-processor Challenge (Not using multi-processor PVM). Other machines to be included later are Sun (probably Solaris 2.4), HP and IBM w/stations. Potentially 10 or so machines could be used, but so far the programs have been tested on only 3 of 4 to demonstrate that it works and is worth doing. -- Chris Dean, Email: C.E.Dean@Daresbury.ac.uk DRAL, Daresbury Laboratory, Warrington, Tel : 01925 603353 WA4 4AD, Int : +44 1925 603353 U.K. Fax : 01925 603230 ------------------------------------------------------------------------------- From: frutos@mac.cie.uva.es (Javier de Frutos) Yes, I am using PVM today. Mainly my own FORTRAN codes for numerical simulation of PDEs. Up to now, a net of SUN SPARC stations. J. de Frutos ------------------------------------------------------------------------------- From: "Rainer Sinkwitz" > > Are you using PVM today? No more. (But I know of colleagues at ABB Power Generation using it (responses follow in parentheses"()") > > What are your application(s)? Particle simulation (Computational fluid dynamics) > > How many and what kind of machines have you used with PVM? Sun, SGI, IBM RS6000 (Don't know) > > Any comments and suggestions for improvement are welcome. > Please email the information to pvm-survey@cs.utk.edu. > You might pass over the following to Vaidy Sunderam - he might remember me from the heterogeneous network computing workshop in October 1991 in Tallahassee. I have finished my Ph.D. thesis in summer 1993 where I had been using PVM for real-time interactive particle simulations using PVM 2.4.3. Since then I never worked in the area of cluster computation again. Since summer 1993 I am now working for ABB (Asea Brown Boveri, which builds Power Plants, electrical transmission, and railways) in the Swiss corporate research center working on HCI and Control system engineering topics. I am still interested in the cluster computing area - so please keep me on your list. My email address is now sinkwitz@chcrc.abb.ch. Please remove "sinkwitz@ifi.unizh.ch". Best regards, Rainer ---- oooooo oooo Rainer Sinkwitz sinkwitz@chcrc.abb.ch $ $ $ " ABB Corporate Research VOICE +41-56-768303 $"$$ """"$ CH-5405 Baden-Da"ttwil FAX +41-56-767365 o$o "$o $ooo" Switzerland ------------------------------------------------------------------------------- From: coupez@reykjavik.cma.fr (Thierry COUPEZ) |> Are you using PVM today? Yes |> What are your application(s)? Computational fluid and solid mechanic. Forming process simulation by F.E. methods. Meshing and remeshing. |> How many and what kind of machines have you used with PVM? Cluster of etherneted workstation (RS6K). SP1, SP2 ( 8 proc.) Best wishes, Thierry Coupez e-mail : coupez@cemef.cma.fr fax : (33) 93 65 43 04 phone : (33) 93 95 75 75 CEMEF, Ecole des Mines, BP 207, 06904 SOPHIA ANTIPOLIS CEDEX, France ____________________________________________________________________ ------------------------------------------------------------------------------- From: fergal@erin1.ucd.ie (Fergal Mc Carthy) > >Are you using PVM today? Yes, though not at a very high priority at the moment. > >What are your application(s)? I am working with the Advanced Computational Research Group in the Centre for Condensed Matter and Biomaterials, part of the Chemistry department in University College Dublin. Our area of interest is the application of parallelism to the simulation of polymers (primary interest), DNA, and proteins, allowing complex simulations to be run in reasonable time. I myself have only just started recently, and as such have no particular area from which to draw examples. However over the summer I was working with DEC in Galway, and while there I worked on the development of a framework Resource Manager for PVM, PVMRM. I hope to be able to continue working on this application, to tidy up some of the code, improve it's genericity, and improve the default scheduling algorithms. A version of the RM has been used in Platform's LSF V2 release. > >How many and what kind of machines have you used with PVM? Here in the ACRG we have 3 DEC ALpha ws's, 4 SGI Iris Indigo's, and a DECstation 5000 connected to a MasPar 1, all of which have been used in conjunction, connected with PVM, on a number of occaisions. > >Any comments and suggestions for improvement are welcome. Well I have already suggested a number of possible improvements and corrections to Robert Manchek and Jim Pruyne while I was working on the RM over the summer, but a quick summary would be: Using an RM - the console code won't recognise external RM's (this was up to 3.3.4... I haven't checked the latest sources) the unclean pvm_addhost() external RM interface: it doesn't check the return value, just uses it as a count of the number of tid's in msg, so if -1 is returned it attempts to unpack. The SM Protocol: no way to remove a host using the protocol no way to extract task status (running, msg wait, PVM wait) from the protocol no way to exchange state information with PVM or other RM's on startup or handoff, information such as the notify database. no way to modify state information such as host to htid mappings, and tid to host mappings which would be useful for the implementation of task and host migration. (Ok so this is probably a pipe dream!) Fergal -- / Fergal Mc Carthy , | Fergal Mc Carthy, \ | Advanced Computational Research Group, | 31 Fosterbrook, | | Chemistry Department., U.C.D., | Stillorgan Road, | | Belfield, Dublin 4, Ireland. | Dublin 4, Ireland. | \ Ph: +353-1-7062418 | Ph: +353-1-2601304 / ------------------------------------------------------------------------------- From: Luc Timmermans > >Are you using PVM today? Yes, although I only just started using it. > >What are your application(s)? Large-eddy simulation of turbulent flow in channels or pipes. > >How many and what kind of machines have you used with PVM? > I did some preliminary work on a Cray T3D. In the near future I will also use a Convex SPP1000 and (perhaps) a IBM SP2. ========================================================================= | Luc Timmermans | | | Laboratory for | email: L.Timmermans@WbMt.TUDelft.NL | | Aero and Hydrodynamics | phone: +31-15-78 2997 | | Delft University of Technology | fax: +31-15-78 2947 | | 2628 AL Delft The Netherlands | | ========================================================================= ------------------------------------------------------------------------------- From: >> Are you using PVM today? Yes. >> What are your application(s)? use of PVM to facilitate parallelized architecture for compute-intensive graphics (rendering etc); use of PVM as an extended transport layer in a services-based IPC harness for integration into all our software. >> How many and what kind of machines have you used with PVM? HP735 workstations, pentium PCs, NeXTstations; up to a max of around 20 at peak in a hererogenous configuraton. Will add some SGIs at some point. PS - I'm impressed with PVM and so far we have experienced little difficulty using it. Gavin Ferris Senior Software Engineer Cambridge Animation Systems ------------------------------------------------------------------------------- From: roy@dokir.eng.mcmaster.ca (roy underhill) | |Are you using PVM today? Yes | |What are your application(s)? Finite Element Methods | |How many and what kind of machines have you used with PVM? 4 SGI's and 16 suns Best wishes to you folk too! Roy ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ W.R.C. Underhill, Ph.D. email: Trefoil@McMaster.CA Principal, Trefoil Analysis voice: (905) 627-2786 30 Bertram Drive, fax: (905) 627-5930 Dundas, Ontario CANADA L9H 4S9 ------------------------------------------------------------------------------- From: I am using PVM today Computational fluid mechanics Silicon Graphics Iris Indigo, Dec Alpha, Cray T3D ------------------------------------------------------------------------------- From: Knut Eckstein > Are you using PVM today? Yes. > What are your application(s)? a) Parallel Solver for sparse systems of linear equations b) ugp, parallel versions of the (U)nstructured (G)rids multigrid package > How many and what kind of machines have you used with PVM? 7 SGI Indy R4400/R4600 9 i486DX33 and DX2/66 PCs running Linux 1 PC running OS/2 -- Knut Eckstein GKKS, Universitaet Stuttgart WORK: ISD, Pfaffenwaldring 27, 70569 Stuttgart, 0711/685-20 53, FRG HOME: Forststrasse 132, 70193 Stuttgart, 0711/63 98 23, FRG EMail: eckstein@isd.uni-stuttgart.de ------------------------------------------------------------------------------- From: Samir Matar > > Are you using PVM today? YES > > What are your application(s)? > NUMERICAL LINEAR ALGEBRA > How many and what kind of machines have you used with PVM? > PARAMID with *8* Nodes ------------------------------------------------------------------------------- From: > Are you using PVM today? > No - But have used it to good effect and will again. > What are your application(s)? > Simulation work in psychological modelling > How many and what kind of machines have you used with PVM? > Max of 10, but typically 3. Sun Sparc 10s, Decstations, Nexts. --- Dennis Norris MRC Applied Psychology Unit 15 Chaucer Road Cambridge CB2 2EF UK email dennis.norris@mrc-apu.cam.ac.uk Phone +44 1223 355294 x760 ------------------------------------------------------------------------------- From: Alexander Rausch > Are you using PVM today? Yes, currently I am experimenting with pvm 3.3.6. I have problems with getting the exported environment variables via PVM_EXPORT if I spawn a process on a SUNMP architecture. No problem on SUN4, SUN4SOL2. Please fix this. I am programming a workaround in the meantime. > What are your application(s)? I can address a couple of robots from each machine I like > How many and what kind of machines have you used with PVM? SUN4,SUN4SOL2,HPPA,SUNMP (in future LINUX). about 10 machines ------------------------------------------------------------------------------- From: lquaresm@softpar.edinfor.pt (Luis Quaresma (Estagiario do TDTCI)) > Are you using PVM today? > Yes, almost every day I work in this particular project, actualy. > What are your application(s)? > The project I am working on is an European colaboration project (ESPRITT) that aims at developing a parallel C++ implementation over PVM (and MPI). As a test to this new language (and as I work for the portuguese power company) we plan to implement a parallel model of a power plant, and try to run it in real time. > How many and what kind of machines have you used with PVM? > I have now the system up and running on a SUN sparc with Solaris and a HP, I intend to add an IBM RISC 9000 as soon as I have my login there, and by the end of the month a Parsytec parallel machine with 8 processing nodes. > Any comments and suggestions for improvement are welcome. > Please email the information to pvm-survey@cs.utk.edu. I have some questions on PVM: 1 - If I choose to receive a only messages with a certain tag or coming from a certain tid, what happens to the messages that arrive in the meantime? Are they lost or are they read later after a message with the desired tag arrives? 2 - Is there a PVM mailing list? I know there is comp.parallel.pvm, but I prefer mailing lists to newsgroups. Now that sockets are available for windows and windows for workgroups is ready to use, should it not be time for windows version of PVM? Here in EDP, and I think that is the general case in most companies I know, the main computing power are the hundreds of PCs around the building, while there are comparatively few workstations available. Eventually we could install Linux on those PCs, but somehow I don't think that the network manager would see that with good eyes :-) best regards Luis Quaresma ------------------------------------------------------------------------------- From: Alexander Rausch > Are you using PVM today? Yes, currently I am experimenting with pvm 3.3.6. I have problems with getting the exported environment variables via PVM_EXPORT if I spawn a process on a SUNMP architecture. No problem on SUN4, SUN4SOL2. Please fix this. I am programming a workaround in the meantime. > What are your application(s)? I can address a couple of robots from each machine I like > How many and what kind of machines have you used with PVM? SUN4,SUN4SOL2,HPPA,SUNMP (in future LINUX). about 10 machines > Any comments and suggestions for improvement are welcome. > Please email the information to pvm-survey@cs.utk.edu. > Your help is much appreciated. > Best wishes, The PVM Project Team > ps. Previous survey results are available through World Wide Web > at http://www.epm.ornl.gov/pvm/ > pps. We apologize if you received multiple copies of this > message. Our first attempt at sending the message uncovered a > bug in our mailer. Yours Alexander -- ------------------------------------------------------------------------------- Willi Alexander Rausch Universitaet Stuttgart Institut fuer Parallele und Verteilte Hoechstleistungsrechner (IPVR) Breitwiesenstrasse 20-22 D-70565 Stuttgart Germany Tel. : (0711) 7816 247 Fax : (0711) 7816 250 EMail: rausch@informatik.uni-stuttgart.de talk : rausch@leonardo.informatik.uni-stuttgart.de ------------------------------------------------------------------------------- ------------------------------------------------------------------------------- From: Claudio Gallo > I am currently using PVM and I have tried both master-and-slave technique and SPMD. I work on numerical models using finite element and implicit time discretizations on trasport equations. I think it's working fine. Anyway I wonder if it were possible to work using non-blocking receive. I'm not an expertise in this matter, but I have found some problems when I use SPMD mode and I exchange informations between among slaves: a non-blocking receive could be useful I think, because it would avoid some bottlenecks. Also, if it were possible to have a proper strategy and a set of subroutine to properly measure CPU time, overhead, and latencys, well, it would be great! Anyway, I greatly appreciate your work, 'cause PVM is enough easy to understand and quite direct to use. I hope you'll continue this way. Cheers Claudio Gallo ______________________________________________________________________________ - CRS4 - Centre for Advanced Studies, Research and Development in Sardinia --- ------------- Environmental Modelling Group ---------------------------------- ______________________________________________________________________________ Office: Via Nazario Sauro, 10 09123 Cagliari, Italy Tel. (39 70) 2796 280 Fax (39 70) 2796 301 e-mail: fender@crs4.it ------------------------------------------------------------------------------- From: Are you using PVM today? Yes, but we cannot link applications running on OS/2 and on Unix What are your application(s)? Scientific data processing How many and what kind of machines have you used with PVM? PC-i486 under OS/2 2.1 SPARC10 under Solaris 2 *********************************************************************** Emmanuel-S. Cohen-Laroque Dept. of Physiology Voice : +41-22-702-53-89 Medical School Fax : +41-22-702.54.02 1, rue Michel Servet E-mail: cohenl@medsun.unige.ch CH-1211 Geneva 4 Swtizerland *********************************************************************** ------------------------------------------------------------------------------- From: jmg@iscte1.iscte.pt (Jose Guimaraes) Dear Colleague, > Are you using PVM today? I have installed pvm and I'm now getting familiar with it. I'm planning to start very shortly an application development using pvm. > What are your application(s)? My application will be a simulator of several scenarious where multiple accesses to multiple information providers are needed. > How many and what kind of machines have you used with PVM? At present it is installed in the network with access via at least 12 SUN4 machines. Shortly will be installed in 2 Silicon Graphics machines (Indy and Indigo 2). Hope this information can help. Best regards, Jose ---------------------------------------------- Research Center: ADETTI -- Associacao para o Desenvolvimento das Telecomunicacoes e Tecnicas de Informatica University: ISCTE -- Instituto Superior de Ciencias do Trabalho e Empresa ADETTI / ISCTE Av. das Forcas Armadas Phone: +351.1.7935000 Ext.: 256 Edificio ISCTE Fax: +351.1.7935300 1600 LISBOA PORTUGAL Email: jmg@iscte.pt ------------------------------------------------------------------------------- From: gwk@crmunich0.cray.com (Georg-Wilhelm Koltermann) > > Are you using PVM today? Several people at this site are using PVM on a regular, frequent basis. > > What are your application(s)? They are porting locally developed applications to a multiple/parallel machine model. > > How many and what kind of machines have you used with PVM? Workstations, Cray mainframes, Cray MPP (T3D). ------------------------------------------------------------------------------- From: Are you using PVM today? YES. What are your application(s)? HARMONIC BALANCE (DETERMINATION OF PERIODIC SOLUTIONS OF NONLINEAR DE) FOR SIMULATION OF ANALOG CIRCUITS. How many and what kind of machines have you used with PVM? 10 RS6000, 10 SUN4/75 Carsten Bergemann bergemann@math.uni-hamburg.de ------------------------------------------------------------------------------- From: lothar@ki6.informatik.uni-hamburg.de (Lothar Hotz) Are you using PVM today? Yes What are your application(s)? Search Tools for AI. Development of a programming language for AI on distributed machines. How many and what kind of machines have you used with PVM? 12 SUN4 4 SUN4SOL2 Any comments and suggestions for improvement are welcome. Native implementation for the CM-5 would be useful for me. lothar Lothar Hotz FB Informatik Universitaet Hamburg Vogt-Koelln-Str. 30 D-22527 Hamburg Germany Tel.: +49 (40) 54715 605 Fax: +49 (40) 54715 572 E-mail: lothar@ki6.informatik.uni-hamburg.de ------------------------------------------------------------------------------- From: angelt@ac.upc.es (Angel Toribio) > Are you using PVM today? YES > > What are your application(s)? Parallelization of a package to manage Binary Decision Diagrams. > > How many and what kind of machines have you used with PVM? SUN, DEC-ALPHA, CONVEX If you need more details please contact me and I'll be pleased to answer. Best regards. -- Angel Toribio o o o Dep. d'Arquitectura de Computadors o o o Universitat Politecnica de Catalunya o o o Keep our planet healthy: Gran Capita s/n, Mod. D6 reduce, reuse and recycle 08071 Barcelona SPAIN U P C ------------------------------------------------------------------------------- From: bernini@cluster.cise.it (Cristiano Bernini) > Are you using PVM today? > Yes. > What are your application(s)? > CFD : combustion + compressible Navier-Stokes + Finite Volume > How many and what kind of machines have you used with PVM? > A cluster of 5 IBM RS6000s Best Regards Cristiano Bernini ********************************************************* * * * Cristiano Bernini * * * * Affiliation: CISE SpA * * Divisione Sistemi e Modelli * * Sezione Matematica Applicata * * Office_Address: via Reggio Emilia n. 39 * * 20090 Segrate, Milano (Italia) * * Office_Phone: +39 2 21 67 22 40 - fax: 26 20 * * e-mail: berni203@alliant.cise.it * * * ********************************************************* ------------------------------------------------------------------------------- From: Jose Damian Lopez Hello, I recieved your email and these are may answer by order: 1.- Yes, I am working with pvm. 2.- My applications are about parallel-distributed aplications on fluids dynamics and ocean models. 3.- My programs run on Convex,HP and linux-486 See you.. -- Madrid, 09 de January de 1995, 11 : 31 ***************************************** * Jose Damian Lopez Maldonado * * Clima Maritimo-Puertos del Estado * * c/Avda. del Partenon 10 * * 28042 Madrid * * Phn: 34 1 524 5500 ext 1627 * * 34 1 524 5568 * * Fax: 34 1 524 5506 * * 34 1 524 5502 * * * * e-mail:jdamian@puertos.es * * http://www.puertos.es * * * ***************************************** ------------------------------------------------------------------------------- From: matt@macsch.com (Matt Jackson) > Are you using PVM today? Yes > What are your application(s)? finite-element analysis > How many and what kind of machines have you used with PVM? up to 8 simultaneously Sun Sparcstation 1+, 2 HP 715 SGI Indigo IBM RS 6000 Cray YMP-EL IBM SP-1 ------------------------------------------------------------------------------- From: lad@sigmoide.fct.unl.pt (Luis Augusto Duarte) I am using today PVM to study distributed systems in my university. My university is NEW UNIVERSITY OF LISBON an we use it in our programs. The PVM is installed in a ethernet network of pc-computers and others computers. I think that PVM is not reliable enough to program with it quickly. Personaly, I found one bug. It was a variable used several times by pvm calls that losted its value. But I am a student, so I do not knowledge enough to be critic of PVM. But, if you want, I can sen you the program with the bug. ------------------------------------------------------------------------------- From: Walter.Lioen@cwi.nl Dear PVM team, > Are you using PVM today? Yes. > What are your application(s)? We, the Initial Value Problem group of Peter van der Houwen of the numerical mathematics department of CWI, Amsterdam are working on parallel ODE/DAE solvers in a project called `Parallel Codes for Circuit Analysis and Control Engineering'. > How many and what kind of machines have you used with PVM? At present we focus on an implementation on a cluster of locally available workstations. Our typical PVM is homogeneous and consists of Silicon Graphics Indys (100 MHz R4000SC) coupled by an ethernet. We could also use FDDI or even ATM, but we try to get a speed-up using fast machines on a relatively slow network. On a Parsytec GCel/512 we use Homogeneous PVM 1.0$\alpha$ (based on public domain PVM 3.2.6). On an IBM 9076 SP1 we use IBM AIX PVMe Version 1 Release 2 (based on public domain PVM 3.2(.5)). Finally, I tried SGIMP and I even played 1 hour with PVM on a Cray T3D (at present I do not have access to such a machine). > Any comments and suggestions for improvement are welcome. I would love to have full Fortran support e.g. - pvmfstart_pvmd and - pvmfgettmask - pvmfsettmask - the functionality of pvmtev.h (Could parameters and statement functions using MIL-STD-1753 bit field manipulation intrinsic functions do the trick? OK, not portable. Using External functions instead of macros? Performance does not seem to be the issue here.) The last category would be for creating a decent xpvm-aware spmd executable. A minor point perhaps: the possibility to do a quiet pvm_mytid. This is for a PVM executable that does its own pvm_start_pvmd, if required. I do not want to hassle users by bogus warnings: libpvm [pid1213]: /tmp/pvmd.232: No such file or directory libpvm [pid1213]: /tmp/pvmd.232: No such file or directory libpvm [pid1213]: /tmp/pvmd.232: No such file or directory libpvm [pid1213]: pvm_mytid(): Can't contact local daemon if the `local daemon' is started by the executable itself. > Your help is much appreciated. Thanks for all your nice work. > Best wishes, > The PVM Project Team Kind regards, __ /o \ Walter M. Lioen (_ \ CWI, P.O. Box 94079, 1090 GB Amsterdam, The Netherlands. / /\ \ e-mail: walter@cwi.nl phone: +31 20 5924101 /_/ \_\ fax: +31 20 5924199 telex: 12571 mactr nl ------------------------------------------------------------------------------- From: "David Ewing" Yes, I am using PVM today. My application is rotorcraft simulation. 8 Silicon Graphics Indy workstations. ------------------------------------------------------------------------------- From: Rob Davies Are you using PVM today? Yes. What are your application(s)? I teach others to use PVM (plus parallel computing in general). Applications I have have most to do with: genetic algorithms, numerical modeling (particle code), fine grained matrix solvers, Computational Number Theory. How many and what kind of machines have you used with PVM? 16 SUN workstations max. nCUBE2 (using a version to allow progs written in PVM to compile/run). ------------------------------------------------------------------------------- From: P.J.Parsons@computer-science.hull.ac.uk (Peter Parsons) > >Are you using PVM today? Yes. >What are your application(s)? Numerical calculations and graphics apps. >How many and what kind of machines have you used with PVM? Sun Sparc - up to 16 SG Indy - up to 4 Transtech Paramid - up to 8 i860 nodes ------------------------------------------------------------------------------- Peter Parsons - Department of Computer Science, University of Hull E-Mail P.J.Parsons@dcs.hull.ac.uk Voice: 44 1482 465016 URL http://web.dcs.hull.ac.uk/~pjp/pjp.html Fax: 44 1482 466666 ------------------------------------------------------------------------------- From: Ralph Caruso We have been trying to use pvm to link two different thermal-hydraulic computer codes together to perform nuclear reactor transient and accident analyses. The codes that we are trying to link are RELAP5 and CONTAIN. ACtually, the work has been going on at the Idaho National Engineering Laboratory and the Brookhaven National Laboratory under contract to us, for about two years. The link portion seems to work well, but we have been experiencing problems with the physics of the two codes, as well as time-step consistency. We believe that pvm could be very useful to us, and we would also like to link some of out thermal- hydraulics codes to our neutron physics codes. Ralph Caruso ------------------------------------------------------------------------------- From: Soeren Frandsen Dear Project Team > > Are you using PVM today? yes > > What are your application(s)? Data reduction of multiple data set. A main job despatches data sets to a set of clients, which perform the reduction and returns the results to the main program, which collects them and stores them. > > How many and what kind of machines have you used with PVM? On SGI indy'es and Power Challenge. On CONVEX 3840 and on an 8 processor SPP machine. > > Any comments and suggestions for improvement are welcome. > Please email the information to pvm-survey@cs.utk.edu. One of the main problems has been to measure performance. Some guidelines for doing that would be valuable. Best wishes Soeren Frandsen Institute for Physics and Astronomy Aarhus University Universitetsparken DK-8000 Aarhus C, Denmark tlph: +45 89423612 Fax: +45 86120740 Email: srf@obs.aau.dk or sfrandsen@solar.stanford.edu ------------------------------------------------------------------------------- From: Uli Bartling Are you using PVM today? 'Today' in the literally sense: NO 'Today' in the sense of 'still using it regularly': YES What are your application(s)? Genetic Algorithm - We place different (sub-)populations on different processors. Every once in a while individuals migrate from one population to another one thus exchanging information. - Currently, for some time consuming fitness functions we are implementing a distributed evaluation of that fitness function in that for every individual one processor performs the evaluation. How many and what kind of machines have you used with PVM? Mainly SUN workstations. Since last week, we have access to an EXPLORER (?) system via a SUN workstation as front end. EXPLORER is a parallel computer based on the Power PC chip for computation and transputers for communication. It's a Parsytec computer. Hope that helps. If you are interested in more details, please let me know. -- ************************************************************************** /\___/\ Ulrich Bartling / \ c/o Gesellschaft fuer Mathematik und Datenverarbeitung (GMD) | * * | German National Research Center for Computer Science (GMD) \ ^ / Schloss Birlinghoven, P.O.Box 1316, D-53754 St.Augustin \_~^~_/ Email: ulrich.bartling@gmd.de Tel.: + 2241 14-2466 ************************************************************************** ------------------------------------------------------------------------------- From: mgs@rio.cos.ufrj.br (Marcio Goncalves da Silva) Dear Colleagues, here are the answers with wich I can contribute to the improvement of PVM. >Are you using PVM today? Yes. >What are your application(s)? I am writing a little part of a program with the student Andre' Belllieny (bellieny@rio.cos.ufrj.br). This software is a tool for evaluating overhead brought by the memory consistency model "release consistency". Each PVM task simulates a hardware architecture that implements a consistency protocol. >How many and what kind of machines have you used with PVM? We have used only SUN Sparcstations2 and have had some problems mounting the virtual machines. Andre' Bellieny has more specific information on this matter. >Any comments and suggestions for improvement are welcome. I will probably write more parts of the program or even other ones, and then I will probably have real suggestions to give. Best Wishes, Marcio Goncalves da Silva ------------------------------------------------------------------------------- From: Geordie +1865 740 011 Are you using PVM today? Yes. What are your application(s)? Genetic likelihoods - ie. recursive computations of probabilities involving heavy use of CPU with only occasional file accesses. Run times of serial versions on DEC ALphas between 1 second and months. How many and what kind of machines have you used with PVM? DEC Alpha, DEC Ultrix and SUN Solaris systems. Maximum configuration of 6 machines simultaneously. Main suggestion: [1] better debugging facilities. [2] a note about system overheads in the documentation - ie. something that would allow us to assess the trade-off between sending (i) lots of small data packets on demand or (ii) occasional large ones and hoping the scheduling is roughly optimal. Alan Y. ------------------------------------------------------------------------------- From: "Ulrich Albrecht" > Are you using PVM today? Yes. > What are your application(s)? I'm working on a distributed robot-simulator, running a viewer-modul on a Personal Iris, a dynamic-simulation on a Sparc 10 and a pathplanning modul on another Sparc 10. Other modules are planned. > How many and what kind of machines have you used with PVM? Personal Iris (irix 4.0.5), Sparc 10 (SunOS 5.2), i80386 running LINUX, i80386 running IBM OS/2 (using Jan Ftacnik's port). Most of the development is done on a single LINUX-Machine or a single Iris, though, to avoid blocking other users. Thanks a lot for your work, Ulrich Albrecht ------------------------------------------------------------------------------- From: pvangool@dutlsb3.lr.tudelft.nl (Paul van Gool) Are you using PVM today? Yes, I am. What are your application(s)? Simulation of aircraft and helicopters. Usual multiple machines allows me to run 'real-time' on my workstation. This speeds up my research. How many and what kind of machines have you used with PVM? 2 x HP9000/735 1 x SGI Iris Indigo 2 x Sun Sparc 10 Keep up the good work. PVM really help me out when I want to have some fast results of my simulation. Paul -- When a distinguished but elderly scientist states that something is possible, he is almost certainly right. When he states that something is impossible, he is very probably wrong. -- Arthur C. Clarke +--------------------------------------------------------------------------+ | internet : P.vanGool@LR.TUDelft.NL Faculty of Aerospace Engineering | | compuserve : 100407,3063 Delft University of Technology | | phone : +31-15-785312 Kluyverweg 1, 2629 HS Delft | | fax : +31-15-781822 The Netherlands | +--------------------------------------------------------------------------+ ------------------------------------------------------------------------------- From: H C Pumphrey > > Are you using PVM today? I use it often and might use it this afternoon, I haven't decided yet. > What are your application(s)? Nonlinear retrievals of water vapour concentration in the stratosphere using measurements from the Microwave Limb Sounder (MLS) on the Upper Atmosphere Research Satellite (UARS). > How many and what kind of machines have you used with PVM? > All Sun machines, Sparcstations I, IPX, 5, 10 running SunOS and Solaris. > Any comments and suggestions for improvement are welcome. (1) A history mechanism on the console. I hate pressing the UP arrow and getting ^[[A instead of the previous command. (2) I often get pvm system errors when I start up or try to add new machines. Usually I find that a pvm daemon has started on the slave machine and failed to communicate this to the master. If I kill the daemon by hand, I can then add the slave. This fault is not because I have some text echoed by my .bashrc file. ------------------------------------------------------------------------------- From: David White > Are you using PVM today? Yes. > What are your application(s)? We are using PVM as a basic message-passing facility for a large image processing project, with multiple executables. > How many and what kind of machines have you used with PVM? We run it on a network of Suns (Sparcs mainly, one Sun 4) with a CM-5 Hope this helps. ------------------------------------------------------------------------------- From: R Abdullah > > Are you using PVM today? Yes. > > What are your application(s)? > Solving large systems of linear equations. > How many and what kind of machines have you used with PVM? Dec-Alphas (only 2 at the moment) ------------------------------------------------------------------------------- Following message from P.J. Spence (jules@rrpmg.demon.co.uk) in response to PVM survey request received by me, D R Crooke, presumably as a NA-Net subscriber as 41179@ibmmail.com. Julian's response follows: I am using PVM My application is FORGE3 - a 3D Finite Element forging simulation (rigid-viscoplastic - i.e. incompressible fluid formulation) using a direct solver (currently). My work has involved testing the solver (in isolation) on Silicon Graphics machines; concentrating on using our Challenge machine as the shared memory makes communications fast enough for speed up. The way forward is to integrate the solver into the program and distribute the data and sparse matrix. I have found that the shared memory PVM on SiG tends to fall over, without warning, when other users (e.g. Patran3) vie for it - would it be possible for pvm progs to wait for such contention to finish (and not crash) : also to clean up the ipcs resources after themselves. The use of the tmp directory for pvm is irritating as (when it fills up) other progs have problems - perhaps an option when running pvm that enabled specification of its workspace? Finally when running parallel programs is it possible easily to run something like dbx on slave processes or is there some other way of de-bugging these codes? Generally pvm is OK the things I have mentioned are more to do with putting programs into production (and the hoary problem of de-bugging). Hope I have given you enough information - please contact me if you wish to discuss my use of pvm. Julian Spence ============================================================================ Whilst this information is given in good faith, based on the latest information available to Rolls-Royce plc, no warranty or representation is given concerning such information, which must not be taken as establishing any contractual or other commitment binding upon Rolls-Royce plc, or any of its subsidiary or associated companies. ------------------------------------------------------------------------------- From: "Y.F.Hu" > Are you using PVM today? Yes > > What are your application(s)? CFD, Numerical Analysis > Content-Length: 256 > > How many and what kind of machines have you used with PVM? > 2. Intel i860 and Cray T3D. ------------------------------------------------------------------------------- From: Alexandra Wagner Hi everyone, > > Dear Colleague, > In order to improve PVM we are trying to understand how > people are using the system. Your feedback is important. > Could you take a moment to answer three questions. Sure > > Are you using PVM today? Yes, I am. > > What are your application(s)? I am working on a parallel version of heuristic state-space search strategies like the A*-algorithm (--> arrtificial intelligence). > > How many and what kind of machines have you used with PVM? I am using a LAN network of SUN4 SPARC workstations. > > Any comments and suggestions for improvement are welcome. I can't think of anything right now but if I do so I will email you. -- Bye Alexandra ---------------------------------------------------------------- Alexandra Wagner (wagner@informatik.uni-freiburg.de) Institut fuer Informatik Universitaet Freiburg Germany ---------------------------------------------------------------- ------------------------------------------------------------------------------- From: Darren P Rodohan Hi! Are you using PVM today? Yes, I have been using PVM for about a year. What are your application(s)? We have written a distributed implementation of the Finite Difference Time Domain (FDTD) method (3 dimensional) for electromagnetic applications. The majority of the parallel computation part has been completed. The code was written in C and used the SPMD computation model with overlapping regions at the boundaries. We are also considering using PVM in our practical exercises for our parallel computing courses. How many and what kind of machines have you used with PVM? We have been using a combination of 8 SUN iPX, iPC and classic workstations. We intend to also use a network of DEC alphas with FDDI in the near future. Comments and Suggestions? The FDTD project is being funded by the Higher Education Funding Councils of England, Scotland and Wales as part of a larger project to exploit spare capacity on workstation networks. There are about 8 other projects that are currently using PVM. I believe that the HPC centre at Southampton are currently constructing a WWW page with details of all ofthe projects. We are currently using version 2.4.2 having read several reports that indicated that the performance was superior to version 3.x. Keep up the good work. Best Wishes Darren. ***************************************************************** * Dr. Darren Rodohan * * Dept. of Electrical Engineering and Electronics * * Brunel University * * Uxbridge, Middx UB8 3PH * * England * * Tel. +44 (895) 274000 ext. 2840 * * Fax. +44 (895) 258728 * ***************************************************************** ------------------------------------------------------------------------------- From: Joachim.Hansson@studorg.luth.se (Joakim Hansson) > >Are you using PVM today? Yes. >What are your application(s)? I am working with pvm in order to construct a software which calculate the difference in speckle patterns between two images (speckle photography). The goal is to get a system which can show tensions and small movements solid object. How a fluids behaves and similar things. The calculations is rather, how should I put it? Heavy? The work is a part of my exam and I have not developed the program from scratch. The major part of the calculations has been imported from a program running under a DOS based system. But I, and my companion, has ported the program to UNIX and delveloped it to a massive parallel working program. Of course the language in use has been C (GNU c to be exact). We have been able to cut the time spent computing from 3 hours (66 mh 486) to about 15 minutes. In some cases even better results have been obtained. >How many and what kind of machines have you used with PVM? In the largest installation I have used 25 sun stations of various kinds. I have also tested installations with mixed environment (HP and sun). SOme problems arised when using this large installation. I have not yet been able to deduce if it is my fault, the network or bugs in the used software (The operating systems is for example different). >Any comments and suggestions for improvement are welcome. >Please email the information to pvm-survey@cs.utk.edu. Well give me some time... ;-) >Your help is much appreciated. > >Best wishes, >The PVM Project Team > >ps. Previous survey results are available through World Wide Web at > http://www.epm.ornl.gov/pvm/ ============================================================================= !tkl-armuo@studorg.luth.se ! ! !frazze@ludd.luth.se ! Blow your brain, smoke gunpowder! ! !p92-jhn@sm.luth.se ! ! !============================================================================ ------------------------------------------------------------------------------- From: Giacomo Fiumara 1. yes 2. development of a parallel code for Ab Initio Molecular Dynamics parallel codes for integral theories of the liquid state brownian dynamics classical molecular dynamics 3. my codes run on a cluster consisting of 6 IBM RISC 6000/mod. 580 as well as on a heterogeneous parallel machine consisting of 4 HP/PA 9000 mod 700 computers, 2 IBM 6000 mod. 550 and 320 and 2 DEC MIPS. Best regards, Giacomo Fiumara (fiumara@vulcano.unime.it) ------------------------------------------------------------------------------- From: vincent@kheops.cray.com (Vincent LE BARAZER) >>> >>> Are you using PVM today? >>> yes. >>> >>> What are your application(s)? >>> MPP applications. >>> >>> How many and what kind of machines have you used with PVM? >>> Benchmarks on Cray's T3D and C90/J90. ------------------------------------------------------------------------------- From: mvklink@WI.LeidenUniv.NL (M.H.W.vanKlink) Hello, here are my answers. > Are you using PVM today? Yes, version 3.2.6 > What are your application(s)? I'm building a parallel raytracer > How many and what kind of machines have you used with PVM? 17 SGI Indy 3 HP 9000 15 Sun Sparc Greetings, Michel van Klink ------------------------------------------------------------------------------- From: Song Koh > > Are you using PVM today? No not at this time. I have several application which use PVM but they have not been used for six months. I am planning to resurrect them in the next several months. > What are your application(s)? Modelling semiconductor devices. > How many and what kind of machines have you used with PVM? Three. DEC RISC, HP 9000/380, Sun Sparc > Any comments and suggestions for improvement are welcome. > Please email the information to pvm-survey@cs.utk.edu. ------------------------------------------------------------------------ Song Koh Harris Semiconductor swk@mlb.semi.harris.com MS 62-022 Voice: (407)-724-7085 P. O Box 883 (US mail) Fax: (407)-729-4960 Palm Bay Road (UPS) Melbourne, FL 32902-0883 ------------------------------------------------------------------------ ------------------------------------------------------------------------------- From: Adam K L WONG > > Are you using PVM today? It had been used as a tool for a postgraduate course (Parallel Computing) in our department. > What are your application(s)? Differnet assignment were set by using pvm as a tool in solving computational tasks aiming for speedup of execution time. eg. Fast Matrix Multiplier. > > How many and what kind of machines have you used with PVM? > The machines we used for the building up of the PVM were mostly workstations runing Unix. > Any comments and suggestions for improvement are welcome. > Please email the information to pvm-survey@cs.utk.edu. This software package is easy to use; with good documentation; BUT with NOT enough application examples ! cheers... -- :------------------------------------/\_/\ | Department of Computer Science ( ) | University of Hong Kong \ ; / |------------------------------------/ \ | Name : (Adam) Kai Leung Wong / \ | Internet : aklwong@sunmp.hku.hk ( ) |___________________________________\_____/ ( \ ) / From: I am responding to your survey request. I am not really using pvm today. I recently got a copy of the program installed it and ran pvm_test. However, I do not know as yet how to actually use the program (1 st time user). I knew of the program due to my advisor at college. My applications will hopefully be to run some method of moments type code. The number of machines I will be using is as yet undetermined. (to be determined by the amount of time it requires to finish my runs in a reasonable amount of time.) Most likely 2-6 machines will be used. I have many more at my diposal, almost all of which will be sun sparcs (from ipx to 20). I do not have any suggestions at this time. I hope this is useful. David Perry ============================================================== D.M.Perry, Engineer MS/43 | E-Systems, ECI Division | 1501 72nd Street North | St. Pete, Florida 33710 | -------------------------------------------------------------- E-mail : dmpa@eci-esyst.com | Phone : (813)381-2000 X3468 ============================================================== ------------------------------------------------------------------------------- From: Mark Stiles 3) I use PVM on a cluster of of 8 HP Risc workstations (HPPA). I plan to use it on an IBM SP1 or SP2 in the near future. 2) I have been using PVM as a "baby sitting" program, that is, keeping the eight workstations busy running essentially serial programs. These programs have been quantum Monte Carlo simulations for single electron charging effects, and solution of continuum differential equations for sutdying growth on surfaces. I am now developing a program that will do more fine grained parallelism. The new program will entail doing three dimensional integrations, where the integrand is expensive to compute, but can be computed on a single node. 1) Yes, am using PVM today. -- Mark Stiles National Institute of Standards and Technology Metrology B206 Gaithersburg, MD 20899 Phone: (301) 975-3745 Fax: (301) 926-2746 ------------------------------------------------------------------------------- From: 1. I do not use PVM regularly. 2. The implementation of parallel training algorithms for feedforward neural networks. 3. Sun IPC and a SPARC 2 work stations - 8 machines. ------------------------------------------------------------------------------- From: "Feairheller, Phil" Are you using PVM today? Yes, I have set up PVM on the CompSci departments Sun stations and on a few Linux boxes, and the department is planning on developing a class on Parallelism, clustering, etc. What are your application(s)? We have not developed any applications yet, but as I mentioned, we will be using it for teaching. How many and what kind of machines have you used with PVM? We have 9 Sun SparcClassics, 1 Sun Sparc10 and two IBM 486's running Linux. We have attempted to run PVM on 6 DEC stations, but are have trouble getting rsh to interoperate between Linux and Ultrix. Any suggestions would be appreciated! Hope this helps -Philip pfeairheller@wcupa.edu ------------------------------------------------------------------------------- COMPANY From: zukas@nixie.mitre.org (Tony Zukas) Are you using PVM today? Yes What are your application(s)? Using PVM to manage and distribute a queueing model that estimates airport delays in the National Airspace System (NAS). How many and what kind of machines have you used with PVM? Currently using a network of 3 to 20 Sun Sparc machines (2's, 10's). Possibility for moving to CM5. ------------------------------------------------------------------------------- From: joaquin@nemesis.etsimo.uniovi.es (Joaquin Ordieres Mere) > > Are you using PVM today? YES > > What are your application(s)? STRUCTURAL DESIGN > > How many and what kind of machines have you used with PVM? SGI 440 SGI 4D/35 SGI 4D/20 HP 9000/715 HP 9000/720 HP 9000/720 HP 9000/712 > -- ________________________________________________________________________________ Joaquin B. Ordieres Mere e-mail: joaquin@nemesis.etsimo.uniovi.es Manager of S.I.G. phone: (34-8) 5104272 Professor Fax: (34-8) 5104242 University of Oviedo Spain ________________________________________________________________________________ ------------------------------------------------------------------------------- From: cung van-dat capran >Are you using PVM today? Yes, pvm 3.3.5. with xpvm 1.0.3. >What are your application(s)? Parallel and distributed algorithms such as Branch-and-Bound, sorting, minimum spawning tree, ... We also use PVM for courses in our university. >How many and what kind of machines have you used with PVM? Sun 3/4 HPPA Linux/PC KSR1 IBM SP1 Best wishes for continuing Van-Dat. ------------------------------------------------------------------------------- From: hjenter@stress.er.usgs.gov (Harry Jenter ) > > Are you using PVM today? No, but we are considering using it in the future. > > What are your application(s)? three-dimensional time-dependent coastal ocean circulation modelng, including flow field prediction and particle tracking. > How many and what kind of machines have you used with PVM? We have experimented with O(10) machines from 3 or 4 manufacturers. We have access to O(100) Data General Aviion workstations here. -- Harry L. Jenter hjenter@stress.er.usgs.gov U.S. Geological Survey (703) 648-5916 | FAX (703) 648-5484 Mailstop 430, National Center "The lottery is just a tax on Reston, Virginia 22092 people who are bad at math." ------------------------------------------------------------------------------- From: visser@chem.vu.nl (Olivier Visser) > Are you using PVM today? Yes > What are your application(s)? Quantum Chemistry Code (the Amsterdam Density Functional program) > How many and what kind of machines have you used with PVM? Currently, only the IBM SP1 and a cluster of RS6000 workstations, but we will be running on a number of different platforms in the near future. > Any comments and suggestions for improvement are welcome. In general pvm is working very well for our application, and we have little problems using it. We have a few problems with portablity: - the startup code of vendors-implementation works different from the public domain code - when using a virtual machine consisting of for example CRAY and RS6K machines, a problem arises in communicating integer arrays since CRAY uses 64 bit integers, and ibm uses 32 bit integers. The first problem is that INTEGER8 is not supported, and the second problem is that the integers need to be converted (we would like to use some general type INTEGER, depending on the machine, and let pvm manage the required conversions). Wish list: Portable startup code (also useable for pvm-e etc) A simple solution for the 'integer problem' A fast implementation of the reduce operation A fast multicast operation (use the network multicast option) A simple manner to destroy the output from the kids Best wishes for 1995, Olivier Visser, Theoretical Chemistry, Vrije Universiteit Amsterdam, The Netherlands. ------------------------------------------------------------------------------- From: "John R. Crowe" > Are you using PVM today? Yes. > What are your application(s)? My research on the signal processing side of ultrasound imaging requires that I run compute intensive 2D and 3D correlations. I use PVM to break the jobs over 10-30 machines here at U of Michigan. > How many and what kind of machines have you used with PVM? The max number of machines I have used at once is probably 30. I've used HP 7xx's, Sparc 2, 10 and 20's and IBM RS 6000's. I'm pretty sure that I've had runs that used all three at once. Definitly runs with Sun and HP's. > Comments I've made the move from PVM 3.2 to PVM 3.3, and I'm contemplating MPI. Do you folks support a version of MPI? Thanks, John Crowe crowej@eecs.umich.edu From: Stefan Hahndel Hello, >Are you using PVM today? Yes. >What are your application(s)? Currently I am writing my doctoral thesis about distributed planning algorithms in multi-agent systems. Especially I am working on the problem how to distribute some typical tasks in manufacturing planning and control. Within this work I developed a negotiation-based distributed planning method, that I myself implemented in a simulation environement running on one workstation. When I heard about PVM I gave some of me students the task to do a second test implementation using PVM in order to test whether this approach is useful in practice. We are now using this first prototype for experiments with this planning system and do further improvements. >How many and what kind of machines have you used with PVM? My student used up to 100 HP workstations (HPPA architecture), but mostly we are using smaller configurations around 40 machines, because they are also used by other users ;) But I also tried it on a network containing different architectures (SUN,HPPA and HP300). >Any comments and suggestions for improvement are welcome. It's a great system. I also did a presentation of our planning system on a small conference (MAAMAW'94) using a foreign network consisting of 20 SUN Sparcs without problems. This showed me how portable distributed software. Only the debugging of the distributed implementations is sometimes not so easy and it would be nice when there are proper debugging tools available. One of my students tried xpvm (if I remember correctly), but was not able to run it. But perhaps there is already a good tool available and I do not know about it ? Best wishes and a happy new year, Stefan hahndel Office: Technische Universitaet Muenchen Phone: +49-89-48095-210 Institut fuer Informatik Fax: +49-89-48095-203 Orleansstr.34 Telex: 3457 tumue d D-81667 Muenchen, Germany Room: 236 PGP public key available on request ! IRC: Hast ------------------------------------------------------------------------------- From: Matt Miller > > Are you using PVM today? > Yes. > What are your application(s)? > Computational fluid dynamics (numerical simulation). > How many and what kind of machines have you used with PVM? > Sun SPARC (2), HP700 (3), HP382 (1). ------------------------------------------------------------------------------- From: "Andy.Phippen" Hello With reference to the survey message you sent out, the application I am currently developing is a parallel genetic algorithm, on a network of 3 Sun single processor SPARCS and one multiprocessing SPARC. I am using pvm for the inter population commuication at present, but may develop the uses as the application itself develops. Hope this is of help Cheers Andy ------------------------------------------------------------------------------- From: pvm-survey> Are you using PVM today? Not immediately. But soon I will be porting an application from IBM SP1 EUI to PVM 3.3 for portability issues. pvm-survey> What are your application(s)? An indusrial collaborators' CFD code. Other members of the team are working with molecular dynamics and others work with structural finite elements. pvm-survey> How many and what kind of machines have you used with pvm-survey> PVM? IBM RS6000 cluster (4 running PVM 2) before SP1 (8 nodes) Sun Sparc IPX (run on 4 although 10 available) Silicon Graphics Power series shared memory machine (4 nodes) pvm-survey> Any comments and suggestions for improvement are pvm-survey> welcome. Please email the information to pvm-survey> pvm-survey@cs.utk.edu. I will be interested to see if version 3.3 has improved performance, although it is unlikely that I shall be able to make the results public. Mark Richardson /| /| /| Parallel Applications Centre / | / | / | 2 Venture Road | | | | | | Chilworth Phone: +44 (0)703 760834 | | | | | | Southampton Fax: +44 (0)703 760833 | | | | | | S016 7NP E-Mail: mr@pac.soton.ac.uk | | | | | | United Kingdom | / | / | / |/ |/ |/ ------------------------------------------------------------------------------- From: tbodine@utig.ig.utexas.edu (Tom Bodine) > Are you using PVM today? Yes, but I am only a system manager. I haven't written anything but test programs for PVM. > What are your application(s)? Geophysical structural seismic processing. > > How many and what kind of machines have you used with PVM? > We run PVM on 5 Sun Sparc 10 and 2 Sparc 20's. Most of them run Solaris 1, a few run Solaris 2. > Any comments and suggestions for improvement are welcome. Some one who uses PVM on a daily basis should be tasked with maintaining the FAQ. I fear upgrading PVM since the installation, and the last upgrade were so painful. The installation procedures need improvement. Question: Is the shared memory option for Solaris multi-cpu PVM working yet. ------------------------------------------------------------------------------- From: S.Miller@rus.uni-stuttgart.de (Shannon Miller) > > Are you using PVM today? I personally do not use it. I sometimes obtain it and install it for the benefit of users here of our Intel Paragon system. In that sense, I'm more of an administrator. We log every invocation of pvmd3 and the log shows that 21 different Paragon users at this site have called pvmd3 a total of 617 times between Feb 09 17:26:55 MET 1994 and Mon Jan 09 14:15:35 MET 1995. Hope this helps. Regards, --Shannon +--------------------------------------+ | Mr. Shannon Miller, Customer Support | | Intel Scalable Systems Division +------------------------------+ | Regionales Rechenzentrum der | smiller@rus.uni-stuttgart.de | | Universitaet Stuttgart | or smiller@ssd.intel.com | | Allmandring 30 (* Letters Only) | Tel: +49-711-678-8738 | | 70550 Stuttgart, Germany | FAX: +49-711-678-8363 | +--------------------------------------+------------------------------+ | * Freight: Allmandring 3A, 70569 Stuttgart 80, Germany | +---------------------------------------------------------------------+ ------------------------------------------------------------------------------- From: >Could you take a moment to answer three questions. >Are you using PVM today? No. >What are your application(s)? Aerospace, propulsion system dynamics. >How many and what kind of machines have you used with PVM? Looking into a Harris NightHawk/Sun Sparc10/Dec Vax 3100 combo. >Any comments and suggestions for improvement are welcome. Currently looking at using PVM. Duane Mattern (mattern@ariel.lerc.nasa.gov.us) (216)433-8186 (office) NYMA @ NASA Lewis Research Center, M/S 77-1 (216)433-8643 (FAX) 21000 Brookpark Road, Cleveland, OH 44135 USA ------------------------------------------------------------------------------- From: Jacek Kitowski Hello, Please, find some remarks on PVM: 1. we use PVM every day for at least 2 years for research and educational purposes. We use also other environments like Express, Linda, p4, NX2. 2. PVM is used for dovelopment of parallel (distributed) algorithms for molecular simulation (molecular dynamics, lattice gas, some trials with CFD, parallel FEM, distributed graphics) 3. SUN SPARCstation2, SLC,IPX, 470, 10 IBM RS/6000 320,520H, 550..... HP9000/700 720, 712/80 CONVEX C3220, C3820 experiments were done also with TMC CM5 4. Suggestions: - global operations similar to Express excombine() or NX2 gdsum(), gopf() - more information available via XPVM, like in ParaGprah for example. Sincerely yours ------------------------------------------------------------------ Dr. Jacek Kitowski Institute of Computer Science | email: kito@uci.agh.edu.pl The Stanislaw Staszic | tel/fax: +48 (12) 33-94-06 University of Mining and Metallurgy al. Mickiewicza 30, 30-059 Cracow/Poland ------------------------------------------------------------------ ------------------------------------------------------------------------------- From: I'm not using PVM now. Indirectly, I'll be using it sometime this year. We'll be using the Paradise parallel database system from Univ. of Wisconsin, which will use PVM. We're using a network of Sun's as a parallel machine to start with. Eric Hanson ------------------------------------------------------------------------------- From: gray@ctron.com (Eric Gray) >> Are you using PVM today? Yes although mostly as an academic pursuit. >> How many and what kind of machines have you used with PVM? We have used PVM on IRIX 4.x Indigo and Indie systems, Sun OS Sparcstations and Ultrix DEC workstations. >> Any comments and suggestions for improvement are welcome. >> Please email the information to pvm-survey@cs.utk.edu. We have been able to use PVM in three different parallel programming models; Hierarchical (Client-Agent-Server), Pipe-Line and Data Parallel. In each of these approaches, PVM presents little difficulty in use (design and coding). While PVM is versatile and generally easy to use, there is room for improved debugging capabilities. It would be preferable that changes or improvements to PVM take into account the variety of parallel programming paradigms that may be in use. ------------------------------------------------------------------------------- From: Piero Lanucara Hy, my answer to your questions are: Are you using PVM today? >Yes, PVM 3.3 What are your application(s)? >My actual interest is concerning the Flame Front Propagation problem. >Myself (with M. FALCONE and C. TRUINI, Dip. Matematica Univ. di Roma >"La Sapienza") have presented at the FIRST EUROPEAN PVM USERS' GROUP MEETING >a talk titled: >Parallel algorithms using PVM for the evolution of interfaces How many and what kind of machines have you used with PVM? >We have tested and run this application on our cluster of 8 DEC Alpha 3000/500 >FDDI interconnected via DEC Gigaswitch. Best regards, Piero Lanucara, CASPUR E-Mail lanucara@caspur.it ------------------------------------------------------------------------------- From: djw@gypsy.stetson.edu (Dr. Donna Williams) We have used PVM as a tool in a course on Parallel Computing. It is not currently being used, but will again in the near future. We used it on a network of 9 Sun4 workstations The biggest problem was the /tmp files left behind when a program crashed or was interrupted. Overall we found it an excellent system. Donna Williams Associate Professor Stetson University DeLand, FL 32720 ------------------------------------------------------------------------------- From: Anselmo Lastra > > Are you using PVM today? Yes. > What are your application(s)? A simulator of a parallel graphics computer that we are building. We use a PVM-based system as a software development platform. > > How many and what kind of machines have you used with PVM? > Just a few machines at a time (4 or so). Mostly HPs. Some RS6000s. ------------------------------------------------------------------------------- From: "Mike Griffin" Are you using PVM today? No. I plan to use it starting this summer. What are your application(s)? I'll be doing computational fluid dynamics (3-D Incompressible Navier-Stokes). How many and what kind of machines have you used with PVM? None yet. I will be using about seven or eight SGI Indigo2's and Indy's and one Crimson. ------------------------------------------------------------------------------- From: > > Are you using PVM today? Yes > > What are your application(s)? I don't know, what! I'm programming a mathematical Problem (Genetik and Evolutionary Algorithm for solving linear Problems) > > How many and what kind of machines have you used with PVM? Up to 15 machines of IBM RS6000 > > Any comments and suggestions for improvement are welcome. > Searching for Errorrs at a client is not easy MfG Thomas W"alde Computing Center of University of Karlsruhe Am Zirkel 2 76131 Karlsruhe Tel: +49 721 608-4736 Fax: +49 721 32550 E-Mail: waelde@rz.uni-karlsruhe.de ------------------------------------------------------------------------------- From: Are you using PVM today? Yes. What are your application(s)? I am using it in conjunction with Krylov method of lines solutions of partial differential equations. How many and what kind of machines have you used with PVM? I use it with a network of Sun workstations (1-25). Skip Thompson Radford University Radford, Virginia 24142 ------------------------------------------------------------------------------- From: rrallo@etse.urv.es (Robert Rallo) >Are you using PVM today? YES >What are your application(s)? Neural Networks and Pattern recognition >How many and what kind of machines have you used with PVM? We use a heterogeneous cluster of five UNIX machines: 2 RS6K (560 & 220), 1 HP9000 / 730 1 SUN SparcStation 1 SUN 630MP >Any comments and suggestions for improvement are welcome. >Please email the information to pvm-survey@cs.utk.edu. Yours sincerelly, "Salut (a tots ...) i llarga vida a la Republica de l'ETSE" ******************************************************************** Robert Rallo Escola Tecnica Superior d'Enginyeria (ETSE). Universitat de Tarragona (URV). ******************************************************************** Carretera de Salou s/n. 43006 TARRAGONA (SPAIN) Tel. 34 (9)77 559686 Fax. 34 (9)77 559710 E-Mail: rrallo@etse.urv.es URL: http://www.etse.urv.es/rrallo/Intro.html ******************************************************************** From: Dr Yuyan Wang Are you using PVM today? Yes. What are your application(s)? Computational Fluid Dynamics. How many and what kind of machines have you used with PVM? Four stations which are SUN20,2xSUN4 and SGI-indigo --------------------------------------------------------------------- | Dr. Yuyan Wang | Tel: + 44 225 826826-5593 | | School of Chemical Engineering | Fax: + 44 225 826894 | | University of Bath | E-mail: Y.Wang@uk.ac.bath | | Claverton Down | WWW: | | Bath BA2 7AY UK | http://www.bath.ac.uk/~cesyw | --------------------------------------------------------------------- ------------------------------------------------------------------------------- From: "Hoffman, Peter x3357" We use PVM for a major application involving computational fluid dynamics (CFD). We usua;;lly work with four workstations, but sometimes as many as sixteen, from SGI, SUN, HP, IBM, and DEC. We often put a Cray in the loop as well. Peter Hoffman ------------------------------------------------------------------------------- From: dlander@idss.nwa.com (David Anderson) > > Are you using PVM today? Not at the present time. > What are your application(s)? Varies, but the original intent was to manage data "blobs" of unstructured information. That project hit the bit bucket late in 1994 and we're not sure when it might be resurrected. > How many and what kind of machines have you used with PVM? Our PVM test had six systems, all RS/6000 AIX Version 3.2.3. -- David L. Anderson | Voice: 612.726.0775 | Northwest Airlines, Inc. dlander@idss.nwa.com | Fax: 612.726.0521 | 5101 Northwest Drive, Dept J3750 | | St. Paul, MN 55111-3034 ------------------------------------------------------------------------------- From: edlg@server01.lncc.br (Eduardo Lucio M. Garcia) com prazer acuso recebimento dos seminarios de Matematica Computacional. estava de ferias e nao pude comparecer. desejo continuar recebendo os avisos via e-mail. atenciosamente, eduardo garcia. ------------------------------------------------------------------------------- From: hetu@simmibm4.imi.nrc.ca (Jean-Francois Hetu) Are you using PVM today? No What are your application(s)? CFD applied to material processing (Navier-Stokes , solidification, phase-change, Mould filling, hyperbolic transport equations, ... ) How many and what kind of machines have you used with PVM? 2 IBM 590 1 IBM 550 3 IBM 320h 1 HP 735 1 Sun Sparcstation 10 model 41 1 SGI (Pers. Iris 4D35...) 2 DEC Alpha 1 DEC station 5000 >--------------------------------------------------------------------------- Jean-Francois Hetu Institut des materiaux industriels Email: Jean-Francois.Hetu@nrc.ca CNRC / NRC Tel : (514) 641-5082 75, de Mortagne, Boucherville, QC Fax : (514) 641-5104 Canada J4B 6Y4 ------------------------------------------------------------------------------- From: Florian Fuchs My answers to your questions: > Are you using PVM today? No, but I used it the last two years. > What are your application(s)? It's a Multi-Agent System that I wrote for usage in a production planning system. > How many and what kind of machines have you used with PVM? About 100 HP700 machines. ----------------------- Florian Fuchs ------------------------------------------------------------------------------- From: Enno Cramer Dear PVM Project Team, in respect of your survey I ll give you some information of the project we are using pvm with. On our Institution, the Lehrstuhl fuer Werkzeugmaschinen at the WZL (Werkzeugmaschinenlaboratorium) of the University in Aachen we are using and developing programms for the finite-element calculation and optimization (shape & topology) of mechanical structures (especially of tooling machines). Beside comercial software for FEM calculation we are using the self-developed programsystem STRATHOS with is specialized in the structural and thermal optimization of composites. The main goal of our research project "Parallelization of FEM- Optimization" (founded by the German research fund DFG) is to speed up calculation time of STRATHOS. We first investigated domain-decomposition techniques. To speed up the performance of a single FEM-Analysis we implemented a multi-frontal FEM-equitation solver in STRATHOS using pvm as interface. Because restructuring the knots with the nested-disection method would cost too much time, we are using only two solver proceses in connection with up to eight proceses doing the calculation of the element matrices in parallel. Our hardware consists of a workstationcluster of 10 RS/6000 worstations. Using all 10 workstation, STRATHOS runs nearly 3 times as fast as the sequentiell version, although the other machines are busy running a CAD-System (CAEDS). We hope to improve performance by replacing the file exchange of the various modules by communication over pvm, thus kind of pipelining the optimization process. Our overall experience with pvm ist quite good, although there are some drawbacks your probably already aware of: - I do not know a way of doing or controlling either dynamic or static load balancing - There is no way to controll the amount of data that has been sendet, but not yet received (-> buffer overflow) - It is imposible to measure performance, because the blocking receive operation does not increment the user time counter. So it is not possible to tell how long an operation blocked. I would like to have a monitor tool (which may of cource cost performance) to analyse the overall behavior of the progam system. This is a thing that I cannot realize, because of lack in time and knowledge. If you got to know about such a tool through your survey, please let me know. Bye, Enno H. Cramer CRM@WZL-WM!.WZL.RWTH-AACHEN.DE Lehrstuhl fuer Werkzeugmaschinen am Laboratorium fuer Werkzeugmaschinen und Betriebslehre (WZL) der RWTH Aachen Steinbachstr 53B, 52074 Aachen, Germany Tel.: +49 241 807449 Fax.: +49 241 8888 293 ------------------------------------------------------------------------------- From: Johan Meijdam > Are you using PVM today? Yes, I do > > What are your application(s)? > Building a task graph interface without using a parallel computer > How many and what kind of machines have you used with PVM? > HP Apollo 735 SUN Sparc Classic Best wishes, Johan Meijdam. johan@pa.twi.tudelft.nl ------------------------------------------------------------------------------- From: xw@pip.fpms.ac.be (Xiao Wei) Dear members of The PVM Project Team: First sincerely thank you for providing such a useful parallel utility to us. Then I answer those three qlestions listed in your letter: 1: Of course, I am using PVM today. I will try to use it better in future. 2: My research is in Artificial Neural Networks, I have built my parallel simulation for ANN with PVM. 3: Now in my laboratory, we have a SUN Sparc server 670mp that has two processors, a SUN ELC workstation and a Alpha machine which has 16 processors. These machines link each other with a ethernet. I built my simulation in this environment. In my work, sometimes I was bothered by how to getting accurate transmission time between two projects run on both same and different machines. Maybe I have not found good solution to solve it. Can you give me some good suggestion? Gest regards, WEI Xiao ------------------------------------------------------ Mr. WEI Xiao Address: Rue des Tuileries 4, tel:0032-65-374056 Email:xw@pip.fpms.ac.be ------------------------------------------------------------------------------- From: torque@world.std.com (Torque) We have not yet used PVM. We're thinking of porting our implementation of Linda to run on top of PVM. Hope this is helpful to you. Iain Bason ------------------------------------------------------------------------------- From: tobita@marc.co.jp (Akio Tobita) Dear PVM staff: > In order to improve PVM we are trying to understand how > people are using the system. Your feedback is important. > Could you take a moment to answer three questions. The Answer is yes! > Are you using PVM today? I'm affraid not. That was moths ago. But other colleagues are using that of HardWare vender's version. > What are your application(s)? MARC and Mentat Those are FEM code and its pre/post processor. > How many and what kind of machines have you used with PVM? I used to use 5 HP9K/700 and 3 IBM RS6K. Today's user is using one Multiprocessor platform. > Any comments and suggestions for improvement are welcome. > Please email the information to pvm-survey@cs.utk.edu. I want to ask a question in return. What is happening to the standardization procedure. I guess PVM is a de facto standard right now. Is it possible for PVM to adopt the idea of Virtual Shared Memory, which shares Virtual Data Segment throughout the network? And is it worth doing? Best Regards, -- Akio Tobita Systems Group M M Nippon MARC Co., LTD. MMM MMM 4F Daiichi-Seimei Bld., 2-7-1 Nishi-Shinjuku, Shinjuku-ku MM MMMM MM Tokyo, 163-07 Japan. WWW WWW Tel. +81-3-3345-0181 Fax. +81-3-3345-1529 W W E-mail. tobita@marc.co.jp ------------------------------------------------------------------------------- From: gock@masc39.rice.edu (Mark Steven Gockenbach) > > Are you using PVM today? > Yes > What are your application(s)? > Seismic velocity inversion, with applications to hydrocarbon exploration. > How many and what kind of machines have you used with PVM? > 4 IBM RS6000's ------------------------------------------------------------------------------- From: David Mackay > > Are you using PVM today? Yes. > > What are your application(s)? Benchmark code (computational chemistry) written to use pvm. > > How many and what kind of machines have you used with PVM? One machine, Intel Paragon. ------------------------------------------------------------------------------- From: bala ramakrishnan Dear SColleagues, Yes , I am currently involved in developing distributed computing version of a Navier Stokes flow solver (TLNS3DMB) currently. The application is being run on SGI and SUN workstations. The project is almost at a completion stage. Thats all for now. balasu ------------------------------------------------------------------------------- From: Charles.Fineman@Eng.Sun.COM (Charlie Fineman) I am using PVM. Although I am quite familiar with it, in fact, I have no direct experience in using it. Charlie Fineman ------------------------------------------------------------------------------- From: I am maintaining a large ground water model on the xps5, xps35, and rx machines at ORNL. I don't run this model myself, other than to test it. ------------------------------------------------------------------------------- From: alien@essex.ac.uk (Adrian Clark) > Are you using PVM today? Yes. > What are your application(s)? We are researching into image processing on parallel and distributed systems. In particular, we are currently developing an image processing software package that is capable of serial or parallel use, with PVM being one of the underlying routes to parallelism. This software also uses Tcl/Tk, and we have put together something rather like the PVM console as an extension verb to Tcl; someone out there might be interested. > How many and what kind of machines have you used with > PVM? We have used: ~25 SUN4 5 SGI5 2 HPPA 1 ALPHA 1 LINUX (development platform only; no networked computation) I had to nobble the `arch' script to make it support my flavour (Slackware) of Linux. ..Adrian -- Dr Adrian F Clark PHONE: (+44) 206-872432 (direct) FAX: (+44) 206-872900 Dept ESE, University of Essex, Wivenhoe Park, Colchester, Essex, C04 3SQ, UK. ------------------------------------------------------------------------------- From: mpapakhi@browndwarf.ucs.indiana.edu (Mary Papakhian) > > Are you using PVM today? Yes. > > What are your application(s)? In users' home directories. > > How many and what kind of machines have you used with PVM? One cluster of 15 IBM RS/6000s, models 580 and 370. > -- Mary Papakhian STARRS System Administrator University Computing Services Indiana University (812) 855-2597 ------------------------------------------------------------------------------- From: Xiao Xu Dear Colleague, I use PVM every day recently. We have a cluster composed of 8 Silicon Graphics (6 x R4400 & 2 x R4600). The application is solving 3-D Navier-Stokes equations. The explicit algorithm has been successful, but the implicit algorithm is still in testing. Since the only 8 processors available, the cluster is only used for coding the implicit algorithm. The Cray T3D is also available, which uses PVM, for the implicit algorithm. There are two things that I am not happy. 1. The PVM can not load the jobs according to the loading of the processors, i.e. lack of load banlance. 2. The job will be stopped if one processor have problem. e.g. someone reboot one of the computer, the job will dead. Regards! Dr. Xiao Xu Dept. of Aerospace Eng. Univ. of Glasgow Glasgow G12 8QQ U.K. January 9, 1995 ------------------------------------------------------------------------------- From: We are not actively using PVM today, but we plan to use it in the future. The application is a 3D linear elastic boundary element code. Both the integrations and the solver can be parallelized, but we do not have an out-of-core parallel solver, thus we are not actively using it. We have it running on 4 DEC ALPHA's and 1 DEC Station 5000. Regards, Bruce Carter ------------------------------------------------------------------------------- From: Ricky E Sward > > Are you using PVM today? No. I used it for a class here at the Air Force Institute of Technology last quarter. > > What are your application(s)? In the class, we used PVM to implement mutual exclusion using several different algorithms. THis was part of the class project to learn more about the ways distributed operating systems can implement mutual exclusion. It was interesting. > > How many and what kind of machines have you used with PVM? We used PVM with up to 10 separate SparcStation 20's. They are all part of a larger network of SUN workstations available here at AFIT. > > Any comments and suggestions for improvement are welcome. > Please email the information to pvm-survey@cs.utk.edu. > We used XPVM in an attempt to visualize what was happening on the virtual machine. XPVM was too slow to actually use for debugging or executing test runs. If you could speed up XPVM it would be an excellent tool. PVM as a tool was quite straightforward to use and an excellent distributed o.s. utility. ------------------------------------------------------------------------------- From: zhangq2@burntumber.psc.edu (Qiming Zhang) Are you using PVM today? Yes. What are your application(s)? Materials Science codes. How many and what kind of machines have you used with PVM? Cray C90 and Cray T3D. ------------------------------------------------------------------------------- ------------------------------------------------------------------------------- From: Raphael Finkel application: distributed tree search How many / what kind of machines: Haven't gotten underway yet, but will be on the order of 20 sparc-5 machines. ------------------------------------------------------------------------------- From: seeley@cloud.plh.af.mil (Guy Seeley) not using pvm today no applications 0 machines used I am actively considering using pvm for IR scene modeling. If we decide to move into a parallel mode of operation pvm will be our #1 candidate ------------------------------------------------------------------------------- From: Fred Kulack Excerpts from mail: 6-Jan-95 PVM survey pvm-survey@cs.utk.edu (720) > Are you using PVM today? Yes > What are your application(s)? I'm currently using PVM as strictly an educational and experimental platform for a small project in a Distributed Systems class > How many and what kind of machines have you used with PVM? about 8 machines 4 RS/6000 w/AIX 3.2.5.7, 4 Intel 486 w/Linux. Fred Kulack Team OS/2 is for EVERYONE! - Bill Gates. ;-) ----------------------------------------------------------------------------- Open Systems Enablement (Internal : kulack@rchland ) IBM Rochester, MN (Internet : kulack@vnet.ibm.com ) ph: 507.253.5982 (Tie line : 553.5982 ) WWW Personal Page (Personal Page http://www.rchland.ibm.com/~kulack/) ------------------------------------------------------------------------------- From: ellis@hstc.necsyl.com (Dennis R. Ellis) Could you take a moment to answer three questions. ** Sure, no problem. Are you using PVM today? ** Yes. What are your application(s)? ** Several benchmarks that we are looking at have PVM implementations. ** Some utilities for our supercomputer for which we want real-time ** displays. ** As a message passing learning tool. How many and what kind of machines have you used with PVM? ** NEC SX-3 supercomputer and SGI Personal IRIS. Any comments and suggestions for improvement are welcome. Please email the information to pvm-survey@cs.utk.edu. ** 64-bit integer support would be *VERY* helpful. +-----------------------+--------------------------+--------------------------+ | Dennis R. Ellis | HNSX Supercomputers Inc. |(HNSX Supercomputers is a | | Sr. Benchmark Analyst | 4800 Research Forest Dr. | subsidiary of NEC Corp.) | | Ellis@hstc.necsyl.com | The Woodlands, TX 77381 |All the normal stuff about| | (713) 364-0034 | FAX: (713) 292-8818 | opinions being mine... | +-----------------------+--------------------------+--------------------------+ ------------------------------------------------------------------------------- From: |> Are you using PVM today? Yes. |> What are your application(s)? Scientific: cluster molecular dynamics, eigenvaule of large symmetric matrices... |> How many and what kind of machines have you used with PVM? DEC 5000 and ALPHA; maximum 6. |> Any comments and suggestions for improvement are welcome. Every thing is fine. Probably a more user-friendly version of xpvm will increase the productivity. Sicerely, Mihai Horoi ------------------------------------------------------------------------------- From: "John S. Zelek" > > Are you using PVM today? yes > > What are your application(s)? mobile robot navigation. The perceptual and control processing is dispersed over many processors. > > How many and what kind of machines have you used with PVM? SUN and SGI. > > Any comments and suggestions for improvement are welcome. > Please email the information to pvm-survey@cs.utk.edu. shared memory. -- John S. Zelek Centre for Intelligent Machines (CIM) McGill University, Montreal tel: (514)-483-0780 e-mail: zelek@mcrcim.mcgill.edu www: http://www.cim.mcgill.ca/~zelek/Home.html ------------------------------------------------------------------------------- From: jspeery@sandia.gov (James S. Peery) > > Are you using PVM today? yes - V.3.3 > > What are your application(s)? > Strong shock wave physics and continuum mechanics > How many and what kind of machines have you used with PVM? > SGI, SUN4, HP, SUNMP > Any comments and suggestions for improvement are welcome. > Please email the information to pvm-survey@cs.utk.edu. > We will be moving to MPI as it becomes available. We enjoyed PVM, but it appears to us that MPI will be something to standardize on. Thanks, -James ------------------------------------------------------------------------------- _/_/_/ _/ _/ _/ SANDIA NATIONAL LABORATORIES _/_/_/ _/ _/_/ _/ _/ Department 1431 _/_/ _/_/_/ _/ _/ _/ _/ Albuquerque, NM 87185-0819 _/_/_/_/_/_/ / _/ _/_/ _/ tel: 505 845 9336 _/ _/_/ _/ _/_/_/ _/ _/ _/_/_/_/ fax: 505 844 0918 _/ _/_/ _/ _/_/_/ Computational Physics Research and Development Department James S. Peery - jspeery@sandia.gov ------------------------------------------------------------------------------- From: gwm@hayworth.LANL.GOV (Gregg W. McKinney) PVM Project Team, Yes we are using PVM extensively and have upgraded to PVM 3.3.5. It has become an integral part of MCNP 4A and newer versions. We are in the process of incorporating load-balancing with this capability. Hundreds of MCNP users have taken advantage of this on virtually all Unix platforms, including (to my knowledge) Suns, HPs, IBM RS/6000, Cray (UNICOS), Cray T3D, Meiko, CM-5, SGI. To date the most workstations I have performed timing studies on is 16 (on Suns and IBM RS/6000s). I have also done some studies on the Cray T3D with 128 processors. I have found the PVM software to be one of the most reliable utilities I've used. Continue the good work! Gregg W. McKinney gwm@lanl.gov ------------------------------------------------------------------------------- From: douglas@math.purdue.edu (Jim Douglas) 1. Today: we are in the process of changing workstations to a collection of IBM PPC's. We will install PVM within a few days and begin using it. 2. Applications: Nonlinear pde's in 2D or 3D and time. 3. Previous use: A small family of Sparcs and RS6000's. Imbalance between processors reuced effectiveness. Also, on a 16 processor IBM SP1, where it was very useful. Jim Douglas, Jr. ------------------------------------------------------------------------------- From: sanz@lace01.lerc.nasa.gov (Jose Sanz) Yes, I use PVM now on a routine manner. My application is Aerodynamic Inverse Design and Analysis for Full Engine. I use PVM to run this application on two CRAY C90s. My host is an IBM RS6K-590. I use PVMe on an IBM SP2-160 nodes at NASA Ames. ------------------------------------------------------------------------------- From: Yes I am using PVM today. I am implementing a parallel version of our probabilistic analysis code. Presently, I have implemented Monte Carlo simulations on our workstation network. I plan to implement other probabilisitic methods in parallel soon. I am running exclusively on an HP workstation network (HP 9000/700 series). I have used up to 10 workstations. ------------------------------------------------------------------------------- From: Chris Chase S1A Survey> Are you using PVM today? No. I used it for a particular problem in 1993. I will be returning to that problem in the near future (early 1995). Survey> What are your application(s)? Optimization in the setting of parameter estimation. I use PVM to partition the calculation of the error function. The particular parameter estimation problems that I have been working on deal with data inversion or model matching in remote sensing applications of the earth's upper atmosphere and the magnetosphere. Survey> How many and what kind of machines have you used with PVM? Machine types: SGI Iris4d, HP9000s700 Numbers: I have used PVM using combinations of 8 of these machines at one time. Survey> Any comments and suggestions for improvement are welcome. Survey> Please email the information to pvm-survey@cs.utk.edu. I will be able to make more informed comments later this year after returning to my application that uses PVM. Chris Chase =============================== Bldg 24-E188 The Applied Physics Laboratory The Johns Hopkins University Laurel, MD 20723-6099 (301)953-6000 x8529 chris.chase@jhuapl.edu ------------------------------------------------------------------------------- From: rme@sticky.rockefeller.edu Are you using PVM today? Yes. What are your application(s)? Computing the Karhunen-Loeve transform for online analysis of optical imaging experiments. How many and what kind of machines have you used with PVM? Up to 12 SGI and Sun workstations. Richard Everson. -- Knight Laboratory for Biophysics, Telephone: +1 212 327 8384 Rockefeller University, Fax: +1 212 327 8123 1230 York Avenue, NY, NY, 10021-6399. Email: rme@camelot.rockefeller.edu ------------------------------------------------------------------------------- From: jyluo@chevron.com (Yi Luo) 1. I am not using it today 2. may applications was in seismic data processing. 3. Sun, IBM rs6000, and DEC. ------------------------------------------------------------------------------- From: Are you using PVM today? yes. What are your application(s)? Various. We use PVM as one of several communication back-ends for a run-time system for an HPF-like compiler. We have users that do air quality modeling, image processing, etc. How many and what kind of machines have you used with PVM? DEC Alphas connected with E-net, FDDI, Gigaswitch Sun 4s, e-net Dec PMAX, e-net Intel Paragon ------------------------------------------------------------------------------- From: rkufrin@ncsa.uiuc.edu (Rick Kufrin) > > Are you using PVM today? > Yes. Literally :^) > What are your application(s)? > Two unrelated applications: a molecular dynamics program which is written in a parallel superset of Fortran 77, and a parallel classifier algorithm (a decision rule generator) written in C. > How many and what kind of machines have you used with PVM? > Seven different platforms: Sun, Hewlett-Packard, Silicon Graphics workstations. Cray Y-MP supercomputer. Silicon Graphics Challenge supercomputer. Convex Exemplar (Convex PVM). TMC CM-5 MPP. Rick Kufrin Mathematics and Computer Science Group National Center for Supercomputing Applications ------------------------------------------------------------------------------- From: miner@msssrv X-Mts: smtp I used PVM this fall as part of a Parallel Programming course. I used PVM on SGI, Solaris, and LINUX machines. With the SGIs I used up to 8 machines at once. ----------------------------------------------------------- Jon Miner miner@rapnet.sanders.lockheed.com (My real job) jwm@cs.unh.edu (Graduate work) ------------------------------------------------------------------------------- From: frye@conx.bu.edu (Roger Frye) Reply-to: pvm-survey@cs.utk.edu Are you using PVM today? Yes. What are your application(s)? 1) client/compute-server communication. 2) (intended) portable message passing for migrating various CMMD codes from CM-5 to other parallel supercomputers. How many and what kind of machines have you used with PVM? Have used SGI5,CM5 as a client/server pair. Intend to use CM5, SGI, IBM and CRAY multiprocessing. Any comments and suggestions for improvement are welcome. I was surprised by the poor performance I got by coding a client/server system so that the server blindly sent messages and the client read them when ready. I got much better performance by having the clinet request each message. Perhaps I just need a better model of PVM internals in order to understand this behavior, but it would be nice if the simpler protocol were more efficient. -Roger Frye (617) 353-8277 ------------------------------------------------------------------------------- From: harkin@fubar.cs.montana.edu (Gary Harkin) > > Are you using PVM today? Yes. > > What are your application(s)? Mostly classroom projects. However, we are runnin OSF/2 on an Alpha, so it uses PVM and therefore, we are using it also. > > How many and what kind of machines have you used with PVM? DEC Alpha, HP 9000 Series, DEC Mips, SGI > ------------------------------------------------------------------------------- From: rbarrett@wrangler.LANL.GOV (Richard Barrett) >Are you using PVM today? Yes. >What are your application(s)? General support of PVM throughout LANL. Developing example code. Developing utility routines. >How many and what kind of machines have you used with PVM? IBM rs6000, Suns, HP, DEC, SGI. Cray t3d, CM-5, Intel iPSC Probably others. >Any comments and suggestions for improvement are welcome. You gotta love it. Regards, Richard ------------------------------------------------------------------------------- From: grit@CS.ColoState.EDU (Dale Grit) > > Are you using PVM today? I used it in a course on parallel programming that I taught during the Fall 94 semester. The students wrote one program (Gauss-Jordan Elimination). > > What are your application(s)? See above. > > How many and what kind of machines have you used with PVM? We installed PVM on a set of 8 HP400 workstations. > > Any comments and suggestions for improvement are welcome. Installation was more painful than we had hoped. I had been using PVM on a set of Suns. I had hoped to be able to set up the HP environment also so I could see what types of issues the students would face. But I gave up trying to set up my system to also allow the use of PVM across the HPs. We have a global shared file system across all department machines. When I tried to install the HP version, my Sun version died. It seemed difficult to get the environment variables set up correctly, and the specification of where parts of PVM had to reside in the file system seemed restrictive. The students kept their sense of humor reasonably well. A couple energetic ones started working on their assignments early and provided valuable hints for helping the rest of the students to get their environments/directories set up properly. Dale Grit Computer Science Dept. Colorado State Univ. Ft. Collins, CO 80523 grit@cs.colostate.edu ------------------------------------------------------------------------------- From: Luis Gerardo de la Fraga Are you using PVM today? Yes, I am. What are your application(s)? Three-dimensional recontruction from electron microscopy specimens. How many and what kind of machines have you used with PVM? I'm using a reduced version of Oak ridge PVM for one parallel machine PARAMID (from Transtech Parallel Systems Corp). Cheers Luis Gerardo de la Fraga Centro Nacional de Biotecnologia. 28049 Madrid ------------------------------------------------------------------------------- From: jvz@cyberia.cfdrc.com (Jeroen van der Zijp) > > Are you using PVM today? Yes. > What are your application(s)? Visual Computing Environment for distributed CFD Applications > How many and what kind of machines have you used with PVM? Mostly SGI's; other machines are being used, such as DEC/Alpha, IBM, HP. > Any comments and suggestions for improvement are welcome. > Please email the information to pvm-survey@cs.utk.edu. > Your help is much appreciated. > > Best wishes, > The PVM Project Team > PVM is a very good effort; keep up the good work. It is very much appreciated. Jeroen van der Zijp ------------------------------------------------------------------------------- From: steveb@norwalk.cray.com (Steve Buonincontri) Our customers want to PVM across J90s without any load balancing problems etc. Also, our customers want a shared memory version of PVM on the PVP CRI machines. My personal opinion about the PVM on the PVP machines is that is a potential customer coming off a cluster of workstations tries it, he will probably not buy the PVP from CRI. When I was using it, I could not spawn from the pvm console and I had to start the group server myself. THis is a CRI problem and should not be telling this to you. ------------------------------------------------------------------------------- From: Willem.Vermin@sara.nl (Willem Vermin) > > Are you using PVM today? Yes. > > What are your application(s)? For educational use: to learn other people to use pvm. > > How many and what kind of machines have you used with PVM? SP1 (8 nodes), rs6000-cluster(3 nodes), CRAY C90 (4 processors) SGI Onyx, SGI Indy's (8 stations), and Linux > > Any comments and suggestions for improvement are welcome. > Please email the information to pvm-survey@cs.utk.edu. For first-time users it is sometimes a time-consuming job to get pvm running. A checklist would be helpfull. Greetings, Willem Vermin willem.vermin@sara.nl ------------------------------------------------------------------------------- From: sanmarti@inrets.fr (Jorge San Martin) Currently,i.e we have not used PVM today. We have installed PVM on our SUN stations (Sparc 1 and 2). We hope to install PVM also on a PC using GNU-LINUX. We would like to create some applications which should eventually run on IBM (SP1 with 8 nodes) machines for the study of mechanics of continuous media using the finite element method. We currently have a project (thesis for a French Diplome) on the parellisation of a numerical solver using sparse matrices. The numerical programme was created by the French Highways and Roads Department for aid in modellising problems in the civil engineering domain. Best wishes and good luck for the future to the PVM Project Team. ------------------------------------------------------------------------------- From: Christian.Schaller@lrz-muenchen.de (Christian Schaller) > > Are you using PVM today? Yes, I'm currently preparing an introductory course for pvm at our cite (Leibniz Computing Center) > > What are your application(s)? Mainly test examples and some tutorials, but I'm going to port my waveform relaxation code to pvm. > > How many and what kind of machines have you used with PVM? Very different: I've installed pvm on a cluster of suns a cluster of hps on our cray y-mp and on our parallel computer ksr. Best regards Christian Schaller -------------------------------------------------------------------------------- Christian Schaller email: schaller@lrz-muenchen.de Leibniz-Rechenzentrum phone: +49-89-2105-8771 Barer Str. 21 fax: +49-89-2809460 80333 Munich Germany -------------------------------------------------------------------------------- ------------------------------------------------------------------------------- From: Dave Remien Reply-To: dave@hyperk.com > > Are you using PVM today? Not at the moment. Not doing any programming or applications stuff; more like network compatability junk for a while. > > What are your application(s)? > Primary PVM candidate is MCNP. > How many and what kind of machines have you used with PVM? > HP9000/[78]00, Sun, SGI > Any comments and suggestions for improvement are welcome. > Please email the information to pvm-survey@cs.utk.edu. > I think you're doing a great job; PVM is a valuable tool. Keep up the good work, and I hope that DOE can see it's way straight to keeping you funded. ------------------------------------------------------------------------------- From: "Mary E. (Beth) Allen" On Fri, 6 Jan 1995 pvm-survey@cs.utk.edu wrote: > Are you using PVM today? Yes. > > What are your application(s)? > Very simple, client/server groups. I am just learning PVM. > How many and what kind of machines have you used with PVM? I have only used PVM on our network of about a dozen SUN4s. ------------------------------------------------------------------------------- From: "Guillaume Colin de Verdiere" On Jan 6, 7:51pm, pvm-survey@cs.utk.edu wrote: > Are you using PVM today? > Yes, I'm using pvm 3.3 and the T3D version on a daily basis. > What are your application(s)? > My application is a parallel volume renderer which can run either on a T3D or workstations. > How many and what kind of machines have you used with PVM? > I mainly use Silicon Graphics (Indy, Indigo 2 , Power Chalenge with IRIX 5.2; Onyx 2xR8000 with IRIX 6) and the CRAY T3D. I happened to use SUN4 with SunOS 4. > Any comments and suggestions for improvement are welcome. I tried to use PVM 3.3 with XPVM on SGIMP but, as Jim Kohl told me, instrumentation of SGIMP is still to be done. I wish I could have it for the next release. I'd greatly appreciate to have the same version of PVM on workstation *AND* the CRAY T3D. I do know that's not easy since CRAY wants to have it's own improved version but that would help us a lot. Maybe some common programming efforts could solve this ? >-- End of excerpt from pvm-survey@cs.utk.edu Happy new year to all the team. -- Guillaume Colin de Verdiere CEA CEL-V 94195 Villeneuve St Georges Cedex FRANCE email: coling@limeil.cea.fr or: coling@acl.lanl.gov until August 15 1995 thanks to DGA/DRET/DS (France) ------------------------------------------------------------------------------- From: fsfrick@bones.lerc.nasa.gov (David Fricker) > > Are you using PVM today? Yes. > What are your application(s)? CFD code development. > How many and what kind of machines have you used with PVM? IBM RS/6000; SGI 4D/320, 4D/340, 4D/480, Indigo, Indigo2, Onyx; Sun SPARC2, SPARC10; HP 9000s730, 9000s750. Maximum number in one application: 18. -- -------------------------------------------------------------------------- David Fricker voice: 216-433-5960 NASA Lewis Research Center email: fricker@lerc.nasa.gov I don't speak for any government (and hope I never will) -------------------------------------------------------------------------- "Of the seven dwarves, only Dopey had a shaven face. This should tell us something about the custom of shaving." -- Tom Robbins, ------------------------------------------------------------------------------- From: Yokokawa Mitsuo | Are you using PVM today? Yes. | What are your application(s)? A numerical simulation on the Rayleigh-Benard convection by using the direct simulation Monte Carlo method | How many and what kind of machines have you used withPVM? I am using PVM with up to 32 processors on IBM SP2 at Cornell Theory Center. Mitsuo YOKOKAWA Visiting Scientist at Cornell Theory Center on leave from Japan Atomic Energy Research Institute E-mail: yokokawa@sugar.tokai.jaeri.go.jp ------------------------------------------------------------------------------- From: mcqueen@MCQUEEN.CIMS.NYU.EDU (David M. McQueen) > Are you using PVM today? Yes. (That is, yes we are using PVM currently, not literally today.) > What are your application(s)? We solve the Incompressible Navier-Stokes equations in the presence of immersed elastic fluids, primarily for biological fluid dynamics problems. We are currently studying flow in 3D computer models of the human heart and of the Virchow-Robin spaces in the human brain. The numerical solution of the N-S equations is via a finite-difference approach on a 128x128x128 computational grid. > How many and what kind of machines have you used with PVM? IBM RS-6000 workstations in virtual machines consisting of 8 nodes and 64 nodes. ------------------------------------------------------------------------------- From: Per Loetstedt Answers to your questions: 1. We are using PVM today on a regular basis. 2. Computational Fluid Dynamics. 3. 2-10 Silicon Graphics R4000 machines in a cluster. Per Lotstedt Saab Military Aircraft Linkoping, Sweden ------------------------------------------------------------------------------- From: ridout@shasta.plk.af.mil (Brian Ridout (WSAS)) > Are you using PVM today? Sorry We are not currently using PVM. > > What are your application(s)? When we begin using PVM it will be for assorted solid 3D modeling codes. > > How many and what kind of machines have you used with PVM? The machines will be Sun Sparc stations 2's, 10's, and 20's. > The installation was great. Very easy. Thanks for the quality. ******************************************************************** * Brian Ridout (Phillips Laboratory) (voice) 505-846-4349 * * PL/WSAS (DSN) 246-4349 * * 3550 Aberdeen Ave SE (FAX) 505-846-4374 * * Kirtland AFB, NM 87117-5776 (E-mail) ridout@plk.af.mil * ******************************************************************** ------------------------------------------------------------------------------- From: sei@masc5.rice.edu (Alain Sei) PVM Team, Please find below my answers to your PVM survey. > > Are you using PVM today? > No Iam not at the moment. > > What are your application(s)? > Wave propagation simulations for seismic problems. > > How many and what kind of machines have you used with PVM? > 3. Sun, IBM RS6000, Cray Sincerely, Alain Sei ===================================================== | Department of Computational & Applied Mathematics | | Rice University | | Houston, Texas 77251-1892 | | | | E-mail : sei@rice.edu | | Phone: (713) 527 4805 x 2288 | ===================================================== ------------------------------------------------------------------------------- From: alfieri@dg-rtp.dg.com (Robert Alfieri) > > Are you using PVM today? > No, but we may port it to our systems. > What are your application(s)? Commercial multiprocessing applications on large SMP/NUMA machines and on clusters of UNIX servers. It would be nice if a user could configure his PVM application so that part of it runs within a NUMA machine (with appropriate CPU affinity) and part runs across a cluster of such machines. The runtime decision would depend on the availability of machine resources and on the desired degree of high availability. If it doesn't already, it would be nice if PVM were cluster-aware and had some hooks related to high availability (e.g., initiating recovery after node failure). There are no standards in the clusters space. There is an opportunity for someone to define a standard way of doing message-passing, distributed locking, and application start/stop/recovery within a cluster environment. If the standard also runs on a NUMA without changing application code, even better. > > How many and what kind of machines have you used with PVM? > Have never used it and have never looked at the interfaces. ------------------------------------------------------------------------------- From: arunp@engr.uky.edu (Arun P. Muthukumar) > >Are you using PVM today? yes > >What are your application(s)? experimental programs electromagnetics algorithm on Sun's > >How many and what kind of machines have you used with PVM? Sun's, HP's, Next, and Convex Exemplar > >Any comments and suggestions for improvement are welcome. it will be helpful to have a manual or more documentation for XPVM > > arun ---------------------------------------------------------------------------- Arun P Muthukumar phone : 606-252-1089 Dept.of EE. E-mail : arunp@engr.uky.edu Univ.of Kentucky Lexington. WWW-URL: www.engr.uky.edu/~arunp ---------------------------------------------------------------------------- ------------------------------------------------------------------------------- From: Gentelmen While I am not using PVM today, we do use it regularly on a large (270 workstation) SUN network for horrendous problems in number theory that take a LOT of parallel crunching. A splendid piece of software - keep up the good work!!!! Dr Simon J Shepherd University of Bradford, UK. ------------------------------------------------------------------------------- From: courtox@univ-rennes1.fr (Francis Courteaux) 1) Yes I'm using PVM today 2) teacher in University in parallel systems IBM System Engineer working with customer using IBM clusters and IBM SP2 Application: Sismic, hydraulic, Ray tracing 3) In University we have many machines so PVM is currently ruuning at Nantes University with a cluster of 8 RISC S/6000 and 8 SUN Sparc2 at IRESTE scholl: 16 RISC S/6000 at Nantes School of mines: 16 SUN Sparc10 With Customers: I have SUN RISC S/6000 SGI and 2 SP2. I use PVMe with SP2 and sometimes PVM/6000 when I have only RISC S/6000 installed. 4) What I want: a native support of ATM, because I need small latency at a small price. Thank you for working for us. I need also good samples with real application. Question: Is your book available ? from MIT press, the name I think is PVM:Parallel Virtual Machine A users's Guide and tutorial for Parallel Computing. ------------------------------------------------------------------------------- From: "Gerald Loeffler" Are you using PVM today? Yes What are your application(s)? Molecular Dynamics of solvated proteins, demo applications for teaching purposes How many and what kind of machines have you used with PVM? 4 HP9000/720 1 SGI Indigo (R4000) 1 SGI Indy (R4000) 1 SGI Indy (R4400) Any comments and suggestions for improvement are welcome. A generic (i.e. object-oriented) implementation in C++ would be great. This would contain classes like Task, Buffer, Group. I tried to write this as a wrapper to the existing API, but this is very difficult, since the C-API uses integer identifiers to objects a lot, which doesn't translate very well into the object-oriented world. A generic C++-implementation (as I said) of PVM would be much cleaner. -- Gerald Loeffler PhD student in Theoretical Biochemistry Email: gl@mdy.univie.ac.at Phone: +43 1 40480 612 Fax: +43 1 4028525 Mail: University of Vienna Institute for Theoretical Chemistry Theoretical Biochemistry Group Waehringerstrasse 17/Parterre A-1090 Wien, Austria #include "fancy_ASCII_picture" #include "funny_statement" #include "standard_disclaimer" ------------------------------------------------------------------------------- From: jlcruz@sun630.uco.es (Jose Luis Cruz Soto) Are you using PVM today? yes What are your application(s)? numerical simulation on fluid mechanic How many and what kind of machines have you used with PVM? 6 machines SPARC-IPX (Sun) + 1 Alpha (Digital) ------------------------------------------------------------------------------- From: emily@cc.gatech.edu (Emily Angerer Crawford) > Are you using PVM today? > Yes. > What are your application(s)? > I am the support person for PVM at Georgia Tech's Office of Information Technology, High Performance Computing group. I have mainly run test cases but have encountered people running all sorts of applications, including mainly some molecular dynamics codes and atmospheric modelling. > How many and what kind of machines have you used with PVM? > Georgia Tech officially has PVM installed on a cluster of three RS/6000's, a Cray YMP-EL, and a KSR-2 (although the last is not installed to utilize multiple processors unless the user's task does so explicitly). We have recently added an SP/2 which will use PVM as its main library for parallelization. I have worked on all of these machines as well as on some unofficial installations on SUN4 architectures. For my personal use, I have installed it on a pair of 486-DX/2-66's running LINUX (and encountered no significant problems in doing so). -- Emily Angerer Crawford OIT High Performance Computing (emily@cc.gatech.edu) ------------------------------------------------------------------------------- From: Isaac Hasbani > > Are you using PVM today? A: Not yet just starting > What are your application(s)? A: Computerized Fluid Dynamics > How many and what kind of machines have you used with PVM? A: Since we are just starting,.. SGI Regards Isaac Hasbani ------------------------------------------------------------------------------- From: I set up PVM to be used for running MCNP (Monte Carlo N-Particle transport code) on multiple SUN4 type workstations. This implementation of MCNP-PVM was set up as a back-up for running MCNP on lower powered desktop workstations (SUN elc, slc, ipx, ... up to SPARC 2 technology) in parallel when the higher powered servers are unavailable. I have used up to six of these workstations at a time. I am not aware of anyone using PVM recently, but with some rummored changes in charge rates for server use, it may come into more extensive use in the near future. Karl Hillesland Westinghouse Hanford Company Richland, WA ------------------------------------------------------------------------------- From: sn01@Lehigh.EDU (SUDHAKAR NETI) PVM is being used by Prof. Hasan Barada on IBM 6000 machines at Lehigh. I am in the process of installing PVM on another group of IBM 6000 machines on Ethernet for large Fluid Flow / Heat Transfer computations. We are having some minor difficulties in getting some of the files to work together; but we hope to overcome that with Prof. Barada's help. Thank you and keep up the good work. S. Neti /\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\//\/\ \ Sudhakar Neti / / Department of Mechanical Engineering and Mechanics \ \ Packard Laboratory _______________ / / 19 Packer Drive West | |\ \ Lehigh University | HAVE |/ / Bethlehem, PA 18015-3085 | A |\ \ Phone 610-758-4117 | NICE |/ / FAX 610-974-6466 / 610-758-6224 | DAY. |\ \ .................................. |______________|/ \/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/ ------------------------------------------------------------------------------- From: PHIL M WILLIAMS Hi, 1. No, we are not using PVM TODAY. Although we do still use PVM. 2. Programs - distributed data visualization system, image processing and molecular modelling. 3. HP-735s. Thanks, Phil Williams Laboratory of Biophysics and Surface Analysis Department of Pharmaceutical Sciences The University of Nottingham. ------------------------------------------------------------------------------- From: rob@fspark.ece.uiuc.edu (Robert Wagner) Are you using PVM today? No, but I have used it in the past. What are your application(s)? Computational Electromagnetics. How many and what kind of machines have you used with PVM? I have used PVM with a network of about 20 workstations, mostly SUN Sparc 2 and SUN Sparc 10 machines, plus a few IBM RS6000 machines. Regards, Robert Wagner ------------------------------------------------------------------------------- From: > > Are you using PVM today? yes > > What are your application(s)? astrophysical hydrodynamics > > How many and what kind of machines have you used with PVM? 3 HP700 > > ------------------------------------------------------------------------------- From: rbw@server01.lncc.br (Ramiro Brito Willmersdorf) > > Are you using PVM today? Yes > What are your application(s)? Finite element solution of incompressible fluid flow, elliptic problems in general. > How many and what kind of machines have you used with PVM? CLusters of sun's and IBM RS6000's, IBM SP1 > Any comments and suggestions for improvement are welcome. Native support for IBM SP1 high-performance (eh ;) switch. IBM's pvme is always out of date. *And* you have to pay for it. > Please email the information to pvm-survey@cs.utk.edu. > > Your help is much appreciated. So is your work, > Best wishes, Same to you! > The PVM Project Team > ------------------------------------------------------------------------------- From: Ian Parsons > > Are you using PVM today? > Yes. > What are your application(s)? Various: numerical analysis, game tree search, biochemical-medical docking problems (monte carlo and searching), and research into templated parallel computations and I/O. > > How many and what kind of machines have you used with PVM? HP, AIX, SUN4, SGIX, ALPHA. Networks of 10-20 workstations typically but we have just had the SP2 installed so I am looking forward to exercising PVM on a "fast" machine :) > > Any comments and suggestions for improvement are welcome. Could there please be some mechanism for polling on a partial subset of connected processes? For example, I have an application where the parent (a manager) sends out work to child processes. Before accepting any new messages from the parent of the manager, the manager must first drain a "reply" message from the child (ie. send work only to idle workers). I can block on a child or all processes or on a message type or all message types. I need to be able to block (not busy wait either!!!) on a *selected* sublist of processes. My current solution (a modification of busy wait) is to insert a usleep after the polling call. The interval of the usleep progressively grows to some predetermined value. After getting a message, the interval is reset to 1. A poll or block on a sublist of processes would be really nice! ============================================================================= Ian Parsons Department of Computing Science email: ian@cs.ualberta.ca University of Alberta phone: (403) 492-5869 Edmonton, Alberta, Canada fax: (403) 492-1071 T6G 2H1 office: 493 Central Academic Building ============================================================================= ------------------------------------------------------------------------------- From: Chun Gong >Are you using PVM today? Not today. >What are your application(s)? I am using pvm to simulate a distributed-memory parallel computer and run some program under single-program multiple-data model. >How many and what kind of machines have you used with PVM? I used to run PVM with 4--8 Sun workstattions, but these days I run PVM on a Cary-T3D paralle computer. ------------------------------------------------------------------------------- From: Mark.Hale@cad.gatech.edu (Mark Hale) > >Are you using PVM today? Yes. >What are your application(s)? We are looking into agent-based applications from a design perspective. Agents have three components: a resource, model, and wrap. PVM provides one communication channel for the Communications layer of the wrap. >How many and what kind of machines have you used with PVM? AIX, SunOS RS6000, Sparc >Any comments and suggestions for improvement are welcome. >Please email the information to pvm-survey@cs.utk.edu. Looks great. Keep up the good work. I see your plans for the future and you appear to be on track. Mark Hale ---------------------------------------------------------------------------- CAE/CAD Laboratory | Aerospace Systems Design Laboratory | Georgia Institute of Technology | There are those who work all day Atlanta, GA 30332 | There are those who dream all day (404) 894-9810 | As for me, I work to fulfill my dreams (404) 894-9812 FAX | Mark.Hale@cad.gatech.edu | http://www.cad.gatech.edu/image | ---------------------------------------------------------------------------- IMAGE: Intelligent Mutlidisciplinary Aircraft Generation Environment DREAMS: Developing Robust Engineering Analysis Models and Specifications ---------------------------------------------------------------------------- ------------------------------------------------------------------------------- From: baer@cs.nps.navy.mil (Wolfgang Baer) Are you using PVM today? Not regularly, but have tried some tests on networked SUN stations. What are your application(s)? I am lookin for a program standard interface to allow the development of parallel programs which have some capability of being ported accross platforms. I am using 16 to 48 Transputer and PowerPC processors and have developed our own communications harness. This works great but is not compatible with anyone else. I am tired of rewriting the application for every new version of hardware. So my interrest in PVM is as a potential parallel programming standard. I would like to be able to write the same application as a multi task configuration in a single processor or as a multitask program in multiple processors without rewriting a lot. Wolfgang BAer, Naval Postgraduate School ------------------------------------------------------------------------------- From: harvey_werner@MENTORG.COM (Harvey Werner) > > Are you using PVM today? Yes. Here at Mentor Grapics, we are just starting out to use PVM for about a month now. > > What are your application(s)? We are using PVM as part of a remote analog simulation server. > > How many and what kind of machines have you used with PVM? Just 2 - Sun4 and HP700. > > Any comments and suggestions for improvement are welcome. > Please email the information to pvm-survey@cs.utk.edu. Not at this time. Perhaps in the near future. ------------------------------------------------------------------------------- From: Ming Shih =>> Dear Colleague, =>> In order to improve PVM we are trying to understand how =>> people are using the system. Your feedback is important. =>> Could you take a moment to answer three questions. =>> =>> Are you using PVM today? Yes =>> =>> What are your application(s)? Distributed Flow Calculation (CFD) =>> =>> How many and what kind of machines have you used with PVM? 5 SGI =>> =>> Any comments and suggestions for improvement are welcome. =>> Please email the information to pvm-survey@cs.utk.edu. =>> =>> Your help is much appreciated. === ===================================================== | =========== Ming-Hsin Shih,Ph.D. Mississippi State University | | | ( X ) | Fax : (601)325-7692 Engineering Research Center | | =========== Phone: (601)325-8278 2 Research Blvd. | | / \ (601)325-2467 Starkville, MS 39759 | | O 0 shih@ERC.MsState.Edu U.S.A. | === ===================================================== ------------------------------------------------------------------------------- From: rudnei@mat.ufrgs.br (Rudnei D. da Cunha) > > Are you using PVM today? Yes. > > What are your application(s)? Parallel solution of sparse linear systems and N-body computations. > > How many and what kind of machines have you used with PVM? I've used PVM on networks of workstations (Sun, DEC Alpha, HP, SGI), on the IBM SP1 and on the SGI Challenge. ------------------------------------------------------------------------------- From: campbell@igate1.hac.com (David Campbell) We are not currently using PVM but this is more do to lack of time to learn it and port programs than to lack of interest. My feeling is that we should perhaps make the time, as it would ultimately be repayed in faster cfd runs. Regards, Dave Campbell Hughes Aircraft Tucson, AZ campbell@igate1.hac.com ------------------------------------------------------------------------------- From: Laurence Koehn >Are you using PVM today? Yes. >What are your application(s)? Research. Specifically, developing data migration techniques for parallel processors. I'm using PVM to test the techniques without having to use an actual parallel processing machine. >How many and what kind of machines have you used with PVM? I've used up to four SUN4 machines. Laurence Koehn koehn@manip.crhc.uiuc.edu ------------------------------------------------------------------------------- From: Brian Powell From the fingertips of pvm-survey@cs.utk.edu: > > Are you using PVM today? > Yes. > What are your application(s)? > Many, varied scientific applications. Several scientists are using it for their CFD and Q-Chem applications. > How many and what kind of machines have you used with PVM? > We have it on a Cray T3D (Cray's version) and your version on a cluster of 10 DEC Alpha 3000 workstations. -- Brian o-----------------------The Ohio Supercomputer Center-----------------------o | Brian S. Powell bpowell@osc.edu | o-----------------------"My other computer is a CRAY"-----------------------o ------------------------------------------------------------------------------- From: jzh@ornl.gov (Jeff Holmes) > >Are you using PVM today? Yes > >What are your application(s)? Coupled climate models: ocean, atmosphere, ice, terrestrial controlled by flux exchange driver module. > >How many and what kind of machines have you used with PVM? Intel Paragon, IBM RS6000 ------------------------------------------------------------------------------- From: vst@asl1.ikp.uni-bonn.de (Volker Strom) > Are you using PVM today? > Yes > What are your application(s)? > In the german joint project VERBMOBIL (speech analysis, machine translation et al.) PVM now is used as basis for the inter-module communication enviroment ICE; ask Jan Willers Amtrup for more information. > How many and what kind of machines have you used with PVM? > Three SPARC 10 stations. Volker Strom ----------------------------------------------------- Volker Strom Institut fuer Kommunikationsforschung und Phonetik (IKP) Universitaet Bonn Poppelsdorfer Allee 47 D-53115 Bonn, Germany phone +49 (228) 73 56 40 and -56 31 fax +49 (228) 73 56 39 email vst@asl1.ikp.uni-bonn.de ----------------------------------------------------- ------------------------------------------------------------------------------- From: hartung@aea.ruhr-uni-bochum.de (Klaus Hartung) >Are you using PVM today? Yes >What are your application(s)? Signal Processing, Mutimedia-Applications, Psychophysical Measurements >How many and what kind of machines have you used with PVM? 3 SPARC 10, 1 SPARC 20, 4 Indy, 2 Iris Indigo Klaus Hartung Ruhr-Universitaet Bochum Lehrstuhl fuer allgemeine Elektrotechnik und Akustik 44780 Bochum Tel.: 0234/700 5872 ------------------------------------------------------------------------------- From: "Robert P. Goddard" PVM Survey, The following response is copied from my response to the 1993 survey, e-mailed to you on 16 August 1993. Unfortunately, the situation has not changed. No, I am not using PVM today. My potential future PVM application is in simulation of underwater sound. The current version is the Sonar Simulation Toolset. It produces sampled sound as "heard" by a user-specified moving multi-channel active or passive sonar system in a specified ocean containing any number of moving targets (sound reflectors) and sound sources, plus reverberation, ambient noise, and self-noise. The resulting sound is suitable for listening, or for feeding to the back end of a sonar processor (or a computer model of one). It is used for testing and developing sonar systems, training sonar operators, predicting performance, developing tactics, planning experiments, and interpreting measurements. Reference (out of date): Proc. Oceans 89, IEEE Publication 89CH2780-5, pp. 1217-1222. I anticipate using a mix of a dozen or so Sun-4 and NeXT computers, and whatever else is available at that uncertain future date, if and when I use PVM at all. I downloaded PVM 3.0 from netlib@ornl.gov on 27 April 1993, after reading Dongarra et al., "Integrated PVM Framework Supports Heterogeneous Network Computing", Computers In Physics 7 #2, Mar/Apr 1993, 166-175. My motivation was mostly curiosity, to find out whether PVM has the potential to help me satisfy my very substantial computational requirements. Since then, I haven't had time to pursue the question further. Dr. Robert P. Goddard, Senior Physicist Applied Physics Laboratory University of Washington 1013 NE 40th Street Seattle, WA 98105-6698 206-543-1267 Internet: rpg@apl.washington.edu ------------------------------------------------------------------------------- From: "Roar Skalin" > > Are you using PVM today? > Yes > What are your application(s)? Atmospheric models (numerical weather prediction and air pollution) > > How many and what kind of machines have you used with PVM? PVM is one of serveral message-passing systems in our codes. The codes are used by several people on different platforms, but I believe PVM is mainly used on vector-parallel CRAYs, workstations and to a certain extend on CRAY T3D and Intel Paragon (but here SHMEM (T3D) and NX (Paragon) are used most of the time). Best regards, Roar Skaalin -- +-----------------------------------------------------------------------------+ | Roar Skaalin : Phone: +47 73 59 30 45 | | SINTEF Industrial Mathematics : Fax: +47 73 59 29 71 | | N-7034 Trondheim, NORWAY : e-mail: roar.skaalin@sima.sintef.no | +-----------------------------------------------------------------------------+ ------------------------------------------------------------------------------- From: David Jensen Are you using PVM today? --- YES What are your application(s)? --- DoD Intelligence Applications --- Finite Element Analysis --- CFD (Soon?) --- Game Tree Analysis (Soon?) How many and what kind of machines have you used with PVM? --- RS/6000 --- SP1 --- Sun Sparcstations --- DecStation --- Dec Alpha --- SGI (soon I hope) --- CM-5 (soon?) Any comments and suggestions for improvement are welcome. --- MPI compatible interface. (Bletch - ptuuey) --- Deamon-less mode of operation (Concept - Link equivalent of deamon into executable - rsh start of code on other machines) --- Light-weight thread (or task) support for group members. (Maybe its there already, I haven't looked at user guide for several releases) ------------------------------------------------------------------------------- From: evd@nuccard.eushc.org (Ed DiBella) Not using pvm today Use for iterative reconstruction of emission computed tomography data Have used with 7 machines including Sun, RS6K, Dec Alpha, Dec 5100 ------------------------------------------------------------------------------- From: John Snyder > > Are you using PVM today? Yes > > What are your application(s)? I have ported PGS Tensor's Cube Manager Software Suite from the Intel i860/Paragon message passing paradim to PVM. We are currently using the PVM version on 2 SP2's. The following breifly describes our software package. This overview relates to the i860, but the model is the same for the PVM version. CUBE MANAGER OVERVIEW --------------------- The Cube Manager (CM) is a 3D seismic data processing system designed to run on a MIMD parallel supercomputer. It controls the flow of seismic data from tape or disk through a sequence of 2 or 3D seismic processes, which may take place in separate pro- cessing nodes of the parallel computer, returning the results back to tape or disk. The hardware configuration consists of a workstation network connected to the parallel computer through its resource manager which in turn is connected to the individual processing nodes by internal hardware links. The general archi- tectural philosophy of the CM is to try and hide the complexities of the parallel computer from the user and programmer by means of a combination of parallel data structures and a parallel tool- kit. Major software components of the system are: an operator GUI (CMJC for Cube Manager Job Control), a user GUI (CMUI for Cube Manager User Interface), and subsections of the CM which run on the attached workstations, the resource manager, and in the processing nodes. The MIMD parallel supercomputer currently used is the iPSC/860 supplied by the SSD division of Intel Corporation. This machine is configured with up to 128 processor nodes, an internal parallel disc system known as the Concurrent File System (CFS), and 8mm SCSI tape drives. Any subset of compute nodes must be a power of two and is called a "cube" because of the machine's hypercube architec- ture. The CM can be booted in any size cube. It further divides the cube into individual job partitions which may be any size. The CM software is general enough to be adapted to other MIMD parallel computers as well as to the computational requirements of other data-intensive endeavors. The iPSC/860 is interfaced to the network of workstations by ethernet. One workstation in the network is designated the host (or server) for the iPSC/860 and contains system utility soft- ware such as compilers and linkers. A separate dedicated work- station is the system resource manager (SRM) which serves as the gateway to the system. The SRM is provided by Intel and runs a stripped-down version of Unix. Its sole function is to manage cubes of compute nodes on the iPSC/860. There are two special nodes within the iPSC/860 system itself, the Ethernet and service nodes. The Ethernet node is connected to the workstation network and allows network file operations to be performed on the CFS. The service node is a scalar processor which facilitates non-floating point tasks which would otherwise consume a compute node. Typical CM Network Hardware Configuration ----------------------------------------- ----------------- | SUN WORKSTATION | | REMOTE HOST (1) |----- | INTEL SERVER | | ----------- ----------------- |__________|Ethernet | | | |Node | -------------- | ----------- | | | USER SUN (1) | | | SRM (1) |----| INTEL (1) | | WORKSTATION | | ----------- | | -------------- | | | | | | | | | | | | ----------- -------------- | | | USER SUN (2) | | | | WORKSTATION | | Ethernet -------------- | Network | | | |----------------| | | | | ----------- | | |__________|Ethernet | | | | |Node | -------------- | ----------- | | | USER SUN (N) | | | SRM (2) |----| INTEL (2) | | WORKSTATION | | ----------- | | -------------- | | | | | | | | ----------------- | ----------- | SUN WORKSTATION | | | REMOTE HOST (2) |----- | INTEL SERVER | ----------------- The iPSC/860 supports both 3480 IBM-compatible cartridge tape drives and 8mm exabyte tape drives. In general, all Tensor inter- nal data resides on exabyte 8200 or 8500 tapes. The cartridge drives are used for the input and output of external data tapes. Electrostatic plotting is done via third party Geoscan software that rasters data on an Intel node. Seismic programs, known in the CM system as Procedures, generate plot metafiles on the Intel disc. The format of these metafiles is an internal vector drafting format. Geoscan plotting software is then used to raster the plot metafile on the Intel and the operator spools the rastered data to the ap- propriate electrostatic plotter. Data can also be directly dis- played on the users workstation screen using the XPLOT interactive procedure without having to generate a plot metafile. The Intel iPSC/860 features a parallel diskfile system known as the Concurrent File System (CFS). It consists of multiple disk drives which are typically 1.5 Gbytes in capacity. These multiple disk drives are made to look like one very large contiguous disk- file by the CFS software. Individual CFS files are striped across the constituent disks in the file system. Therefore, the user does not have to be concerned on which physical disk their data resides. All data files are readable and writeable by all nodes in the sys- tem. In general, the local diskfile systems resident on one work- station are cross-mounted on all the other workstations on the net- work. This allows a user access to any file on any of the cross- mounted file systems. File system cross-mounting is accomplished by the Network File System (NFS) software. The Intel CFS diskfile system is not, however, cross-mounted with either the workstation NFS network or with the CFS systems on other Intel machines. This means that the Sun workstation user cannot directly access diskfiles on the Intel CFS from his worksta- tion. It also means that if a user wishes to use a CFS dataset in a job on another Intel computer, the dataset must first be trans- ferred to the other Intel CFS by means of tar tape backup or the File Transfer Protocal (FTP) procedure. In order to allow the user a means of managing their own data on the Intel CFS disfile system, there are three procedures; CFSLST, CFSDEL and CFSTAR that allow the user to list, delete and backup files by CM jobs setup via CMUI. The Cube Manager software itself consists of several Unix dae- mons which run in the various hardware components of the iPSC/860. Three of these: the synchronizer, logger, and log formatter run on the service node. The Executive Shell daemon runs in each proces- sing node. It is the part of the Cube Manager which ties processes together into jobs. Two daemons run on the SRM: the linker and loader. Finally, the operator and user interfaces: CMJC and CMUI, the CM kernel itself, a linker, and the data base daemons run on the front end Sun workstation. All of these daemons are in commu- nication with each other and cooperate to interface with the oper- ator and users, build executable code images, initiate job execu- tion, monitor jobs, collect the log messages and present them to the user after job termination. Clearly this is a complicated sys- tem! To understand it, let us "walk" a job through the whole se- quence. The user builds a seismic job flow interactively on his local Sun workstation via CMUI. This job flow can then be either sub- mitted directly by the user from CMUI for interactive execution or given to the Intel operator in the computer room who submits it by means of CMJC. If it involves tape mounts or if it can be run later as a batch job, then the operator route is more convenient. The CM kernel on the Sun receives the job and breaks it down into its components. It looks around on the host Sun disks for old exe- cutables that have the required seismic processes embedded in them. If it can reuse old execuables it does; otherwise, it proceeds to build new executable images with the specified seismic processes for execution in the nodes. So far, everything has happened on the front end Sun network. The CM kernel then loads the executa- ble images, along with the resident Executive Shell into the nodes by using the loader daemon in the SRM. Communication between the Sun host and the SRM is handled by the pair of linker daemons: one in the Sun and the other in the SRM. The CM initializes the log- ger daemon in the service node. At this point everything is ready to run; however, in order to verify the whole software environment, the CM firsts sends a round robin message to the compute nodes and waits for its return. If all is well, the CM starts the job executing. During execution, the two linker, and synchronizer daemons are available to the job for utility functions. Log messages output from the job or the CM are accumulated by the logger daemon. At the finish of job execu- tion, the CM uses the log formatter daemon to sort the log mes- sages into chronological order and write the resulting CM execu- tion log back to the front end Sun disk. Information concerning tape datasets created or modified is kept track of by the data base daemon. Before discussing the various components of the CM in detail, we should first mention "namespaces" which are the glue that holds everything together. A namespace is a tree-type data structure on which is hung information. Like a tree, a namespace has a root where it begins and a series of branches whose intersections are given simple names. The leaves or fruit of the namespace tree are the data names and their values. So a namespace data structure consists of four major items: the tree structure, the intersection names, the data names, and the actual data values. Namespaces can reside in memory as tree structures or be condensed to a form suit- able for storage on disk or for transmission to other hosts or CM nodes. On disk they can be viewed on the screen by use of the utility: "nscat". Namespaces can be concatenated, split, or pruned to make other namespaces as the need may arise. The highest level of the CM is managed by the GUI: CMJC for Cube Manager Job Control. It oversees all of the software that runs in (is "booted" in) an Intel iPSC/860 cube. Usually the CMJC cube is booted in the computer room on the host workstation and is controlled or managed by the computer operator. However, for test- ing or special work, a user may boot his own cube on his worksta- tion which will communicate with the host workstation over the lo- cal network. In this case a CMJM display will appear on his work- station screen and the user will have complete control over all jobs running under his cube. In either case, the cube is described by the "master namespace" which is a concatenation of all the indi- vidual job namespaces that are running under the cube. The CMJC GUI has four classes of functions: Jobs, Nodes, No- tices, and Namespace. The Jobs function allows the operator to: Check, Start, Abort, or Kill individual jobs. Batch jobs are han- dled with this function as well as user jobs that are hung or im- properly terminated. The Nodes function allows the operator to Enable, Disable, or Inquire nodes in the cube. Usually, hardware problems with nodes require the operator to intervene with this function. The Notices functions: Respond and Delete, provide a mechanism for the operator to communicate with jobs requiring some action such as a tape mount or plotting a file on the electrosta- tic plotter. The last function: Namespsace allows the operator to Browse, Set, or Print the master namespace. Browse is a means of navigating through the master namespace and its constituent job components by listing on the screen the data values at each inter- section. The operator (or user) can probe down to any level in order to monitor job progression and performance. The primary tool for the user to interact with the CM is the GUI CMUI. This can be considered a user-orientated window into the cube managed by a CMJC GUI. The user can construct CM jobs from several dozen CM procedures, submit them, abort or kill them, and monitor their performance. The user cannot control individual nodes, respond to messages, or browse other job's namespaces. The job flow is built by either editing an existing job or by selecting procedures from a list of procedures displayed in a window on the lhs of the CMUI display. Each procedure is selected by clicking the mouse on its procedure name. It then appears as a white box in the job build part of the display with the name of the procedure inside. Each procedure is parameterised by clicking the right mouse button on the white box of the procedure. This will simultaneously list the parameters required by the procedure and, below, general documentation for the procedure's function. The user then selects parameters for input by clicking with the left mouse button. Par- ameter description is automatically presented for that parameter in a documentation window. The parameter values are then typed into a parameter input window. A set of red control buttons appear which allow the user to delete, set defaults, and repeat parameters for the procedure. Once all the procedures for a job have been parameterised, the job can be saved to disk under a user-specified name, and then sub- mitted either directly from CMUI or given indirectly to the Intel com- puter operator. Usually jobs with more than one input tape or that produce an output tape are submitted by the Intel computer operator. When a job flow is saved, it is saved as a namespace disc file under a user-specified name. The ".job" filename extension is supplied by (or assumed by) CMUI. All job files are saved in the "/job" sub- directory under the user's project directory. In order to understand the project filesystem we must first digress a bit. Before jobs can be submitted, the user must have a CM user project account in both the workstation "/tensordb" and Intel CFS filesystems. Usually, only CM data will be written to the CFS. However, the "/tensordb/project" directory contains five sub- directories which perform vital CM functions. They are: "/job", "/log", "/db", "/plot", and "/prog". All ".job" files for the pro- ject are are stored in "/job". All CM logger files are returned to the "/log" subdirectory. They have a ".log" filename extension. The "/db" subdirectory contains data base files describing input and output data resident in the "/cfs/project" filesystem. Other data base information stored in the "/db" directory includes CDP grid definition files for 3-D surveys that have a ".grid" file exten- sion and SEGD header files that have a ".sdb" file extension. Plot files created by raster procedures on the Intel will be re- turned to the "/plot" subdirectory. Finally, nonstandard CM pro- cedure objects are stored in the "/prog" subdirectory. If a job generates a plot metafile, this file is stored under the "/cfs/geoscan" directory as a filename which is based on the ".job" filename that created it. Presently, plots are manually queued for plotting by the operator. Programmers and software support personnel typically operate under a personal project name such as /tensordb/david and /cfs/david, while production groups setup a project for a survey and all pro- cessors working on that survey work in the same project directory. Project directories are usually setup and managed by the system ad- ministrator. For example, a job constructed using CMUI that is created for the project: "Olayoil" is given the name "stack". This job outputs a disc dataset called "finalstack" and a tape dataset called "gathers". For this job, the following files will be generated: Directory FileName Format Comments --------- -------- ------ -------- /tensordb/Olayoil/job stack.job namespace Job flow file created by CMUI. /tensordb/Olayoil/log stack.log ascii Runtime log for stack.job /tensordb/Olayoil/db finalstack.cmds namespace Dataset summary file for CFS dataset 'finalstack'. /tensordb/Olayoil/db gathers.cmds namespace Dataset summary file for tape dataset 'gathers'. /cfs/Olayoil finalstack binary Seismic data for dataset 'finalstack' The remaining chapters in the CM Programmer's Guide that follow cover all the items we have alluded to above with our overview in considerable detail. These fall into, roughly, four categories: functional descriptions of the CM software elements, dataset formats, operational instruction, and general philosophy. Chapters 2 (Creating the Namespace Descriptor file), 3 Creating the Run-Time Code), 5 (The Synchronizer and its Services), and 7 (Special Libraries) fall into the first category. Chapter 4 (Data Structures) forms the second category. Chapters 8 (Special Procedures), 9 (Testing, Debugging, and Releasing Procedures), and 12 (Tutorial) make up the instruction or third category. Lastly, Chapters 6 (Namespaces), 10 (Aspects of Parallism), and 11 (Future Directions) provide the philosophical framework of the whole system. We had considered adding a thirteenth Chapter with exercis- es for the student-user but that number was judged to be unlucky and, therefore, a chapter unlikely to be read. Upon reflection, twelve Chapters of detailed description should be plenty! If anyone works their way through the whole Guide and wants more, they can start again at the Overview and will be amazed at what they have forgotten since starting. The Cube Manager (CM) is a 3D seismic data processing system..... > > How many and what kind of machines have you used with PVM? 2 IBM SP2's (390 and 590 nodes) 2 Intel Paragon's (192 nodes each) 8 Intel i860's (64 nodes each) 7 Intel i860's (32 nodes each) > > Any comments and suggestions for improvement are welcome. > Please email the information to pvm-survey@cs.utk.edu. > > Your help is much appreciated. > > Best wishes, > The PVM Project Team > > ps. Previous survey results are available through World Wide Web at > http://www.epm.ornl.gov/pvm/ > > pps. We apologize if you received multiple copies of this message. > Our first attempt at sending the message uncovered a bug in our mailer. > -- --------------------------------------------------------------------------- |John Snyder | You have been awarded a Certification of Merit | |jps@hstn.tensor.pgs.com | for your good work. Now you must destroy it, | | | after all this is a secret organization. | | | ...The Chief (Get Smart) | --------------------------------------------------------------------------- ------------------------------------------------------------------------------- From: Randy Wilson > Are you using PVM today? Not at the present. > What are your application(s)? When I did use PVM, I was using a ray-traced computer graphics program that I had written. In such programs, each pixel of an image must be computed independently by intersecting a ray sent from the 'eye' to the pixel and on into the 3-D 'environment' or 'world'. There can be reflection/refraction and other rays, and the calculations take quite some time (hours for some images). So I had a 'master' program spawn 10 'slaves' on 10 HP-700 workstations. Each slave would ask the master for one line of the image to work on, until all the lines of the image were calculated, at which point the master killed all the slaves and saved out the final image. > How many and what kind of machines have you used with PVM? Like I said above, I used 10 HP Workstations of a few different kinds. Two were HP-718i (I think), and the others were HP-700's. All ran the same binaries, though at different speeds. > Any comments and suggestions for improvement are welcome. > Please email the information to pvm-survey@cs.utk.edu. I thought it was pretty well done as is. > Your help is much appreciated. You're welcome! ------------------------------------------------------------------------------- From: fchism@craysea.cray.com (Frank W. Chism) > Are you using PVM today? Yes, I use PVM on several current projects. > What are your application(s)? I am a benchmarking speciallist at Cray Research. Most of the codes I work on that use PVM are ports of PVM based codes from others. Currently, I have active projects involving PVM based applications performing molecular dynanics, fuild dynamics, and global climate calculations. > How many and what kind of machines have you used with PVM? I use PVM primarily on one of three CRAY T3D. I have used it on CRAY Y-MPs and C90s as well. > Any comments and suggestions for improvement are welcome. PVM should keep track of the number of calls to the various routines and amount of data moved and provide options to display and reset these counts. > Please email the information to pvm-survey@cs.utk.edu. Done! > Your help is much appreciated. You are welcome. Thanks for asking. > > Best wishes, > The PVM Project Team Good bye and good coding, Frank Chism Senior Analyst Benchmarking and Optimization Group Cray Research Western Region ------------------------------------------------------------------------------- From: Manu Konchady Answer 1. Yes, I am using PVM today. Answer 2. We are developing a coupled climate model on heterogeneous systems using PVM. Answer 3. I have used a SUN Sparc 2, DEC 2100 ALPHA, and a Cray C90. Suggestion: Include documentation on how PVM and MPI will work together. ------------------------------------------------------------------------------- From: rainer@kokopelli.lanl.gov (Rainer Hollerbach) Regarding this PVM survey below, _________________________________________________________ Dear Colleague, In order to improve PVM we are trying to understand how people are using the system. Your feedback is important. Could you take a moment to answer three questions. Are you using PVM today? Yes, I have been using PVM pretty much continuously for almost a year now. What are your application(s)? My application is in astrophysical magnetohydro- dynamics. I am solving the induction equation for the magnetic field in the presence of a prescribed fluid flow. I distribute the angular modes of the field over the various processors, and use PVM to implement the coupling between modes. How many and what kind of machines have you used with PVM? Mostly the T3D here at the Los Alamos Advanced Computing Lab, also a cluster of workstations, when I was still in the development and testing stage. Any comments and suggestions for improvement are welcome. Please email the information to pvm-survey@cs.utk.edu. As far as I'm concerned, PVM is just about ideal as it is. I was particularly impressed at how easy it was to port my code from the workstation cluster to the T3D. It took me about half a day, no more, to implement the few changes in the PVM calls. __________________________________________________________ Rainer Hollerbach Institute of Geophysics and Planetary Physics Los Alamos National Laboratory Los Alamos, NM 87545 ------------------------------------------------------------------------------- From: Farantos Stavros Yes we are ucsing PVM in Molecular Dynamics Calculations, mainly with 4-6 HP-9000 workstations. Stavros C. Faranros ------------------------------------------------------------------------------- From: ding@sneffels.phys.nwu.edu (Ding-Chuan Chen) Are you using PVM today? no What are your application(s)? Ultrsonic attenuation in type II superconductor How many and what kind of machines have you used with PVM? 4 SUN4 and 30 HPPA ------------------------------------------------------------------------------- From: Zhengwen Ju > Are you using PVM today? Yes. > What are your application(s)? Distributed 3-D Iterative Image Reconstruction for Quantitative Single-Photon Emission Computed Tomography (SPECT) > How many and what kind of machines have you used with PVM? Up to 9 DECstations or DEC alphas. zhengwen ju ------------------------------------------------------------------------------- From: "Joseph C. Lappa" > > Are you using PVM today? Yes. > What are your application(s)? Sending data from a SGI ONXY/2 to a Cray C90 to do analysis > > How many and what kind of machines have you used with PVM? > Dec Alpha OSF/1 Cray C90 SGI Indy SGI ONYX/2 Dec PMAX ------------------------------------------------------------------------------- From: Don Dazlich Are you using PVM today? Yes What are your application(s)? global shallow water model (PVM code under development) How many and what kind of machines have you used with PVM? HP PA-RISC workstations Next Workstations Cray T3D ------------------------------------------------------------------------------- From: "Mr. Rakesh R. Patel" To: pvm-survey@cs.utk.edu Subject: Re: PVM survey Are you using PVM today? yes. What are your application(s)? particle simulation using fast multipole method. How many and what kind of machines have you used with PVM? IBM RS6000s, 10 machines ------------------------------------------------------------------------------- From: wtk@Myrias.AB.CA (Wayne Karpoff) We use PVM as an internal research tool to understand concepts that we will incorporate into higher level tools (ie. our software products, esp. PAMS). We run on networked Suns and RS6000s, SMP systems, and MPP systems (ours and IBM SP-2's). We play mostly with synthetic applications as we use this as a research tool only. Are you planning to produce a summary of our survey results? -Wayne -------------------------------------------------------- Wayne T. Karpoff, General Manager wtk@myrias.ab.ca Myrias Computer Technologies Inc. 8522 Davies Road (403) 463-1337 Fax (403) 465-0130 Edmonton Alberta -------------------------------------------------------- ------------------------------------------------------------------------------- From: kryukova@cs.caltech.edu (Svetlana A. Kryukova) > > Are you using PVM today? > Yes. > What are your application(s)? > Mostly combinatorial problems. > How many and what kind of machines have you used with PVM? > 4-8 Suns running SunOS ------------------------------------------------------------------------------- From: abbie@alpha.bldr.nist.gov (Agnes O'Gallagher) Are you using PVM today? yes What are your application(s)? i am working on a parallel version of a code developed by John Gary that solves an elastic wave equation. How many and what kind of machines have you used with PVM? cluster of 4 IBM RISC 6000's group of SUN workstations IBM SP2 (woops, this is PVMe) Any comments and suggestions for improvement are welcome. Please email the information to pvm-survey@cs.utk.edu. on both the IBM RISC 6000 cluster and now on the SP2 i have encountered a tendency for either my code or PVM to hang up and not return. while i know it is possible that this could be due to a bug causing a timing error in my program i have a gut-level feeling that it is some interaction of PVM itself with the system. in the case of the RS6000's the problem started on one particular day and persisted for over a month only to disappear completely when the system was taken down for maintenance. but i was, apparently, the only person who encountered any difficulty during that period. i think that i was not aggressive enough in pursuing clues to the problem. instead, i spent the month tweaking my program to see if i could make it run. but i did try reading those pvml.nnnnn logs and found them very unhelpful although they sometimes contained amusing messages like "the mad fools. when will they learn?" i guess it might have been helpful if there was more actual English in those logs. the funny thing is that PVM went up most easily and ran with the fewest problems on the network of SUN workstations where I installed it myself. not sure what that means. ------------------------------------------------------------------------------- From: "Ruth L. Hinkins" Are you using PVM today? last month, probably next month What are your application(s)? automatic differentiation, mag. field comput. How many and what kind of machines have you used with PVM? workstations, mostly Sun Ruth Hinkins rlhinkins@lbl.gov ------------------------------------------------------------------------------- From: >Are you using PVM today? We offer PVM a Parallel Programming Environment for clusters of HP workstations. >What are your application(s)? None. >How many and what kind of machines have you used with PVM? HP 700 series of workstations. >Any comments and suggestions for improvement are welcome. >Please email the information to pvm-survey@cs.utk.edu. We have plans to support Deterministic Replay that would help developers debug their parallel programs by enabling them to replicate the behavior of previous runs. The race condition that exists between pvm_send and pvm_mcast in the current implementation of PVM forces us to trace all messages between tasks. If this race condition is fixed, it would make our implementation of Deterministic Replay for more usable. We feel that supporting asynchronous notification of incoming messages would be a useful feature. Thanks. Ashok Ambati. ------------------------------------------------------------------------------- From: "Jeff Nash" Reply to: RE>PVM survey Dear PVM, I AM currently using PVM. It is, in fact, critical to my work. I am developing a portable, distributed computing environment, with application to modeling experiments in the field of high-temperature plasma spectroscopy. This work will be appearing in a forthcoming article in the Journal of Quantitative Spectroscopy and Radiative Transfer as: "The Development of a Distributed Computing Environment for the Design and Modeling of Plasma Spectroscopy Experiments", by J.K. Nash et al. I am dependent on the ability of PVM for dynamic spawning and process communication on multiple architectures and on its ability to work in an X Windows interrupt-driven environment. We will also be accessing distributed data archives for substantial amounts of atomic physics data, so data transfer rate is an issue. We are currently developing on SUN and SGI workstations. We will shortly be including HP and IBM workstations. I was very pleased at how simple and free of complications it was to install PVM on the SUN and SGI machines. I have not yet attempted data transfer at full scale; I have some concerns, (albeit unjustified) as to data transfer rate. CONCERNS: I also have concern regarding the ability of PVM to establish communications between different machines. First, it seems that when my home directory is NFS mounted, (we have a network of SGI machines sharing disk), that I can only use one SGI in my virtual machine. This may be resolvable in the current PVM implementation - I do not know. Second, on the SGI network, the system message sent out on login contains 'echo' statements which I cannot have removed. This is not the case for my SUN system. The documentation warns of problems with PVM in such cases. I may have those problems since it seems that I can start PVM from the SGI and connect to the SUN, but not vice-versa (I may have this reversed, I can live with this so I haven't documented it yet). Third, I will want PVM to provide the communications for a general data server - i.e. for more than one virtual machine. I think this is possible with the existing system using "groups" but I haven't explored this yet. I have two uncertainties about the ability of PVM to provide such service in a general way: i) will I be able to use "groups" to allow the database to connect and disconnect from multiple virtual machines? ii) will it be necessary to provide an entry in my .hosts file for every such user? I want a general capability, without the need to know the names of users before they connect. Last, I know my system administrators are uncomfortable with the use of the .hosts file. I would prefer some additional way to provide connectivity. I am a physicist with only two years experience with UNIX systems and my project involves many area besides PVM. I have not, therefore fully researched these concerns. It may be that my answers lie in further exploring the "group" capabilities of PVM - or that I need to explore UNIX systems further. Since you requested comments, I have replied! Let me conclude with the point that PVM has worked SO well, with little investment on my part, that I have been content to postpone addressing the above issues. If someone has an opportunity to help answer these questions, I would be most grateful. Jeff Nash Lawrence Livermore National Laboratory 510-422-7255 jknash@llnl.gov ------------------------------------------------------------------------------- From: su@milne.geology.yale.edu (Liqiang Su) >Are you using PVM today? Yes >What are your application(s)? Scientific calculation (seismology) >How many and what kind of machines have you used with PVM? 10 Sun Workstations. (2 Sparc10, 8 Sparc2) ------------------------------------------------------------------------------- From: >> Are you using PVM today? Yes. >> What are your application(s)? Molecular Biology Fast and sensitive protein database searches. >> How many and what kind of machines have you used with PVM? SGI Power Challenge SUN-SPARC Parsytec PowerXplorer, Parsytec PowerGC Meiko CS2 Reinhard Schneider schneider@EMBL-Heidelberg.de ------------------------------------------------------------------------------- From: cscbp@knuth.mtsu.edu (Dr. Chris Pettey) > Are you using PVM today? Yes. > > What are your application(s)? 1. I am in the process of porting a parallel genetic algorithm to pvm. 2. I use pvm to teach MIMD processing in the parallel processing class. > > How many and what kind of machines have you used with PVM? 1. 5 HP workstations 2. 8-10 Vaxstations > > Any comments and suggestions for improvement are welcome. We have a file server that all our workstations use. Therefore, we can't use the HP's and the Vaxstations in the same virtual machine. It would be nice if we could use all of our machines. Is there a debugger for pvm? I have a grad student that is interested in creating a parallel object oriented debugger. Chrisila Pettey ------------------------------------------------------------------------------- From: "H. B. Runesha" Response to your survey: >Are you using PVM today? Yes I am using pvm and I have been using it for a year now. > >What are your application(s)? I am trying to implement a finite element code on PVM,and an optimization code.I am trying to see if pvm could be a cheaper tool for parallel applications. This is a portion of my phd theses , writing applications using pvm >How many and what kind of machines have you used with PVM? I have used more than 10 Sun workstaions (SUN4) I have been successfull connecting machines with different architectures > Comments: I have been having problems: -connecting machines with different architectures or other sites(location) -sending messages processor to processor (not to all) in synchronous communication for more than two processors or when the data are long. I will appreciate your help Thank you H.B. Runesha Old Dominion University ------------------------------------------------------------------------------- From: yjohn@utirc.utoronto.ca (John Ross) Date: Mon, 9 Jan 95 13:29:57 -0500 To: pvm-survey@cs.utk.edu Subject: Re: PVM survey > >Are you using PVM today? > Yes--off and on. >What are your application(s)? > None--for experimental and instructional purposes. >How many and what kind of machines have you used with PVM? > DECstations, Sun 3, Iris, Indigo, KSR/1. ------------------------------------------------------------------------------- From: Humberto Ortiz Zuazaga PVM> Are you using PVM today? No. PVM> What are your application(s)? Prototype program for solving Boolean Satisfiability problems prepared for a DIMACS Challenge. PVM> How many and what kind of machines have you used with PVM? Up to 60 SparcStations, a standalone linux PC for development. Humberto Ortiz Zuazaga zuazaga@ucunix.san.uc.edu ------------------------------------------------------------------------------- From: apon@dubois.fisk.edu (Amy Apon) > Are you using PVM today? Yes. > What are your application(s)? I am running simple applications (LU Decomposition) for the purposes of measuring and developing a queueing network model of a parallel system. > How many and what kind of machines have you used with PVM? Seven Sun SPARC stations running Solaris 2.3. > Any comments and suggestions for improvement are welcome. > Please email the information to pvm-survey@cs.utk.edu. It would be helpful if you put the tech_faq in the .tar file that includes the distribution and initial Readme, or at least some comments in the Readme file about how to get the User's Guide, the FAQ, and other useful information. Cheers, Amy ---------------------------------------------------------------------- Amy W. Apon, Ph.D., Assistant Professor of Computer Science Dept. of Math and Comp. Sci., Fisk University, Nashville, TN 37208 TEL: (615) 329-8671 EMAIL: apon@dubois.fisk.edu FAX: (615) 329-8634 WWW: http://www.fisk.edu/~apon/home.html ------------------------------------------------------------------------------- From: "Stella Hadjimarkos" Are you using PVM today? YES What are your application(s)? COG: A MONTE CARLO COMPUTER PROGRAM DEVELOPED AT LLNL, THAT DOES PARTICLE TRANSPORT. IT IS USED BY PHYSICISTS AND ENGINEERS IN SHIELDING AND CRITICALITY PROBLEMS. How many and what kind of machines have you used with PVM? FOUR UNIX MACHINES, INCLUDING HP, SUN AND SGI WORKSTATIONS. ------------------------------------------------------------------------------- From: Tim Raymund >Are you using PVM today? No. >What are your application(s)? Space physics and remote sensing. >How many and what kind of machines have you used with PVM? About 30 HP 9000/7xx series workstations. >Any comments and suggestions for improvement are welcome. A "self-parallizing" compiler (e.g. one that would do loop unroll, vector math, etc.) would help alot. We have lots of code, but no time to rewrite in the PVM calls. tim. --- T. D. Raymund | phone: 512/835-3833 ARL:UT | fax: 512/835-3259 P.O. BOX 8029 | internet: traymund@arlut.utexas.edu Austin, TX 78713-8029 | MIME welcome! ------------------------------------------------------------------------------- From: volker@haydn.informatik.Uni-Osnabrueck.DE (Volker Schnecke) > Are you using PVM today? > Yes. > What are your application(s)? > Tree Search Algorithms (Optimization Problems) Parallel Genetic Algorithms for VLSI-Design (Floorplanning and Routing in full custom Physical Design of VLSI-Chips) > How many and what kind of machines have you used with PVM? > Up to 14 Sun Sparcs (ELC, IPX, S10, S20) Only for fun a heterogeneous network of Sun, HP and Next Workstations Parsytec MC-2 and GCel-Transputer-Systems with up to 512 Processors (Due to the small memory (4MB) on a single Transputer I have only done the tree search on this machines) Heterogeneous Network with Tasks on 512 Transputers and Tasks on the Frontend-Workstation (Sun Sparc) of the MC-2 and the GCel ------------------------------------------------------------------------------- From: "P.W. Dowd" % Are you using PVM today? yes. % % What are your application(s)? aerospace mainly, CFD type of applications. % % How many and what kind of machines have you used with PVM? lots... spread over wide distances % % Any comments and suggestions for improvement are welcome. % Please email the information to pvm-survey@cs.utk.edu. yes, reduce the latency to initiate message send/receive. we are interested in using PVM with ATM for the bandwidth but also the bandwidth-distance product (supporting geographically distributed computing). The price to build a message is so high that it destroys the benefit of a high-speed network. all in all, pvm is great - it has opened up a whole new area in computing. -- Patrick Dowd +1 (716) 645-2422 x2127 dowd@eng.buffalo.edu ------------------------------------------------------------------------------- From: optim@bellcore.com (Tami Carpenter) Are you using PVM today. Not at the moment, but that's mainly because parallel computing is not the main thrust of my research and I'm not currently working on anything that would warrant parallelization. What are your applications? Stochastic optimization problems that arise in telecommunications. This is multi-scenario optimization, where the scenarios arise from considering different failures in a telecommunication network. How many and what kind of machines have you used with PVM? Usually, we run using three or four SUN workstations, but we've used as many as 8. These workstations are a combination of SPARC2's and SPARC10's. ------------------------------------------------------------------------------- From: Pinakin Kanubhai Patel >Are you using PVM today? NO >What are your application(s)? Contaminant flow and transport through porous media which is expressed as a set of pdes. This is computational expensive, especially for practical problems, and hence parallelism could be helpful. >How many and what kind of machines have you used with PVM? IBM RS/6000s upto 8 of them .. but I didn't get any speedup. ------------------------------------------------------------------------------- From: Jie Zhang I am currently NOT using PVM. My potential application of PVM will be in exploration seismology based on solving acoustic wave equations. ------------------------------------------------------------------------------- From: K P Wang > > Are you using PVM today? > Yes. > What are your application(s)? finite element analysis, finite difference analysis, numerical integrations. developing portable CFD codes. > > How many and what kind of machines have you used with PVM? SGI, SUN, HP, DEC ALPHA, DEC station, PC LINUX. > > Any comments and suggestions for improvement are welcome. > Please email the information to pvm-survey@cs.utk.edu. > Thank you very much. Kai-Ping Wang ------------------------------------------------------------------------------- From: > Are you using PVM today? We are starting > What are your application(s)? Parallel version of our CFD codes for DNS (Direct numerical simulations) and LES (Large Eddy Simulations) of incompressible turbulent flows in cartesian and cylindrical coordinates. > How many and what kind of machines have you used with PVM? We will run on IBM/SP2 and CRAY/T3D Massimiliano Fatica Ph.D. Student Department of Mechanics and Aeronautics University of Rome "La Sapienza" fatica@orlandi3.ing.uniroma1.it ------------------------------------------------------------------------------- From: David Jones 1. Yes, I am using PVM. 2. I am using it for computational molecular biology (protein structure prediction). 3. I am using it on a network of Silicon Graphics machines (Irix 5.2), including a multiprocessor Challenge server. >---------------------------------------------------------------------------< This message was written, produced and executively directed by Dr David Jones Email: jones@bsm.bioc.ucl.ac.uk | JANET: jones@uk.ac.ucl.bioc.bsm Address: Dept. of Biochemistry | Tel: +44 71 387 7050 x3879 and Molecular Biology, University | Fax: +44 71 380 7193 College, London WC1E 6BT, U.K. | Disclaimer: STANDARD > KEYWORDS : OPINIONS MY OWN NOBODY ELSE'S WHATSOEVER ------------------------------------------------------------------------------- From: stoffel@no2sun.cray.com (Jim Stoffel) > > Are you using PVM today? Yes. > > What are your application(s)? Confidence tests for CRI T3D. > > How many and what kind of machines have you used with PVM? I am using CRI's PVM on the T3D. ------------------------------------------------------------------------------- From: sabrams@deslab.mit.edu (S. Abrams) PVM survey Are you using PVM today? A program using PVM was completed in August 1994 as part of a Master's thesis by a student who has since left MIT. The program is still used. What are your application(s)? We had previous produced a system for robustly calculating regions of specified curvature in a Bezier (bivariate parametric) surface using a newly developed interval-based equation solver. To generalize the problem, this was modified to work for B-spline surfaces. The program works by subdividing the B-spline into its constituent Bezier patches, calculating regions for each patch and then recombining them. Each Bezier patch is worked on as an autonomous process, using PVM to coordinate and schedule the processes on multiple machines. How many and what kind of machines have used with PVM? We have run our program using PVM with one to eight machines, all Silicon Graphics workstations, a mixture of Personal IRIS 4D-25, 4D-35, 4D-120, Indigo2, and Indy. If you have any more questions about our work, please contact me by telephone at: (617) 253-7799, or e-mail at: sabrams@deslab.mit.edu. Stephen L. Abrams MIT Ocean Engineering Design Laboratory ------------------------------------------------------------------------------- From: trfaulk@california.sandia.gov (faulkner thomas r) > Are you using PVM today? I have used PVM on a Paragon at a number of times. I also provide support for PVM for other users on the Paragon. > What are your application(s)? Seismic ray-tracing Function optimization testbeds Molecular Dynamics Materials science and Quantum Chemistry > How many and what kind of machines have you used with PVM? The Paragon and SUN, SGI, HP, DEC and IBM workstations. > Any comments and suggestions for improvement are welcome. My main complaint about PVM is that if anything goes wrong I often have to put printf into the daemon to get an idea what is wrong. As such, I have had to become far too much of an expert in the daemon and libraries than I like. -- Tom Faulkner Parallel System Engineer Intel Supercomputers Sandia National Laboratories, California ------------------------------------------------------------------------------- From: mike@doug.med.utah.edu (Mike O'Reilly) I am using PVM today. I use PVM in image reconstruction. For each projection the processing can be done in parallel. I have used up to 16 Suns, running a variety of OS. Mike O'Reilly, Ph.D. Medical Imaging Research Lab Department of Radiology U of Utah. mike@fermi.med.utah.edu ------------------------------------------------------------------------------- From: "Francisco J. Prieto" - Are you using PVM today? Yes - What are your application(s)? We have been working on the development of a medium-long term planning algorithm for electric utilities that takes into account uncertainty in the demand and the availability of water resources. We have to solve a large stochastic optimization problem, and we use some decomposition approaches, very well suited for its solution in a distributed environment. - How many and what kind of machines have you used with PVM? We have used two clusters of workstations: 1 HP 730 and 3 HP 720 3 HP 715/50 Thank you very much for your interest, and for your work. Sincerely, - Francisco J. Prieto ------------------------------------------------------------------------------- From: "HYNEK JIRI - FAV" : : Are you using PVM today? : YES : What are your application(s)? : PURELY ACADEMIC : How many and what kind of machines have you used with PVM? : circa 10, SILICON GRAPHICS WORKSTATIONS : Any comments and suggestions for improvement are welcome. PVM IS GREAT! : Please email the information to pvm-survey@cs.utk.edu. : : Your help is much appreciated. : : Best wishes, : THE SAME TO YOU The PVM Project Team : ------------------------------------------------------------------------------- From: renato@ticam.utexas.edu (renato silva) Dear PVM Project Team I'm using PVM 3.2.6 to solve CFD problems, at this moment we are solving the convection- diffusion equation and begin to implement the Euler's equations. This basis of those application are the distributed version of the GMRES method for solving the linear system of equations. We are using 16 IBM RISC 6000 model 3BT, interconnected by FDDI. Best wishes Renato Silva renato@ticam.utexas.edu rssr@server01.lncc.br ------------------------------------------------------------------------------- From: choiy@sun.eng.wayne.edu (Yongsuk Choi) Hi! I've used PVM for one semester for my project and research. I was very impressed for the easy learning curve of PVM. PVM is a powerful tool for curcurrent programming. But, I feel some of PVM features should be improved. First, communication speed is slow when I use PVM_SEND and PVM_RECV. That means, more you communicate more you will be slow down your program. When I used PVM for an image compression techniques, I measured that the program with PVM is slower than program with normal C language. Since many people like me use PVM for parallel computing platforms, communication speed is critical for those projects. I hope great success in your PVM project. Thank You for PVM Best Wishes, Yongsuk Choi e-mail: choiy@sun.eng.wayne.edu choiy@nova.eng.wayne.edu ------------------------------------------------------------------------------- From: Jean Lorre We used PVM initially, then converted to RPC's because the latter didn't require any delivery of software to our customers, at least that was the excuse our systems people used. The application was to have many platforms refer to a single data base on one sun. It was a spacecraft navigation data base which any user could update or read from. The machines were sun-4 solaris vax-vms alpha-vms sgi and hp I found pvm very easy to use and rpc's impossible to understand. An expert had to be called in to convert to rpc's. I'm an applications guy and take no interest in screwing with the operating system. Thanks for a nice product. Too bat we didn't keep using pvm. ------------------------------------------------------------------------------- From: ncs@cs.rit.edu (Nan C Schaller) > Are you using PVM today? Yes. > What are your application(s)? I have students using it to do master's projects and have students using it to do parallel graphics projects in undergraduate and graduate courses in computer science at Rochester Institute of Technology. > How many and what kind of machines have you used with PVM? PVM version 3.3.6 is installed on: 62 Sun SPARCstation 2s running SunOS 4.1.3_U1 1 Sun 4/380 running SunOS 4.1.3_U1 21 Sun SLCs running SunOS 4.1.3_U1 21 SGI Indys running IRIX 5.2. ____________________________________________________________________________ Nan C. Schaller Phone: +1.716.475.2139 Rochester Institute of Technology Fax: +1.716.475.7100 Computer Science Department Internet: ncs@cs.rit.edu 102 Lomb Memorial Dr. OR ncsics@ritvax.isc.rit.edu Rochester, NY 14623-5608 BITNET: ncsics@ritvax.bitnet ------------------------------------------------------------------------------- From: rssr@server01.lncc.br (Renato Simoes Silva) Dear PVM Project Team I'm using PVM 3.2.6 to solve CFD problems, at this moment we are solving the convection- diffusion equation and begin to implement the Euler's equations. This basis of those application are the distributed version of the GMRES method for solving the linear system of equations. We are using 16 IBM RISC 6000 model 3BT, interconnected by FDDI. Best wishes Renato Silva renato@ticam.utexas.edu rssr@server01.lncc.br ------------------------------------------------------------------------------- From: Are you using PVM today? We are just starting to. I awaited the arrival of a particular graduate student who wanted to continue a particular project. What are your application(s)? Electrical propagation in the heart, toward understanding causes of sudden cardiac death. We will attempt to port a model previously run on a Convex which uses the Microelectronics Center of North Carolina developed Spice like circuit analysis program CAZM where the "devices" are membrane segments and the current-voltage relationships are from the biological literature. These models need to have 100,000 to 1,000,000 or more nodes to be useful and are both memory and computation intensive. How many and what kind of machines have you used with PVM? To get started, 3 DECstation 5000 RISC machines (Ultrix 4.4) 1 DEC Alpha 500X (OSF1 3.0) Ultimately: several other DECstations another Alpha IBM RS6000 Any comments and suggestions for improvement are welcome. It would be helpful to me if you know someone who has ported CAZM to PVM and also how well such programs (i.e. Spice) work in the PVM environment. Jack Buchanan, MSEE, MD Assoc. Prof of Medicine and Biomedical Engineering Director, Memphis Educational Computer Connectivity Alliance University of Tennessee, Memphis 899 Madison Ave, Suite 801 Memphis, TN 38112 buchanan@bme1.utmem.edu buchanan.jack@memphis.va.gov buchanan@mecca.org ------------------------------------------------------------------------------- From: Are you using PVM today? Yes. What are your application(s)? We have parallelized matrix programs like Matrix multiplication, Transpose, Cholesky Factorization, 2D-FFT, etc. and benchmarks from Livermore Kernel, PERFECT, etc. How many and what kind of machines have you used with PVM? We have used SUN Sparcstation 1, SUN Sparcstation 5, SUN Sparcstation 10, SUN Sparcstation LX, SUN 10/2000 multiprocessor, and SGI Challenge multiprocessor. -Mohammed Javeed Zaki -- ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ Mohammed Javeed Zaki email :- zaki@cs.rochester.edu Computer Science Dept. off. :- 716-275-5605 University of Rochester res. :- 716-272-8375 Rochester NY 14627 home add:-239 B Quinby Rd. Rochester NY 14623 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ ------------------------------------------------------------------------------- From: Ian Glendinning > Could you take a moment to answer three questions. Sure. > Are you using PVM today? Yes. > What are your application(s)? The PARKBENCH and GENESIS benchmark suites. > How many and what kind of machines have you used with PVM? Meiko CS-2, Silicon Graphics workstation cluster, Parsytec GC/PP. > Your help is much appreciated. You're welcome. Ian -- Ian Glendinning HPC Centre, Computing Services igl@par.soton.ac.uk University of Southampton, SO17 1BJ, UK Tel: +44 703 592594 WWW URL: http://hpcc.soton.ac.uk/staff/igl.html ------------------------------------------------------------------------------- From: Jian Huang 1. We are using PVM today. 2. Our application is the parallel computation of fluid dynamics. 3. We have about 10 Sun SPARC workstations and 3 IBM RS6000 workstations here. ------------------------------------------------------------------------------- From: Bob Bagwill > Are you using PVM today? Yes > What are your application(s)? Just a demo, running xpvm and xep/mtile. > How many and what kind of machines have you used with PVM? Linux, Sparc/SunOS 4.1.2, OSF386, Sparc/Solaris 2.3, Digital Alpha/OSF1 > Any comments and suggestions for improvement are welcome. I don't know the system well enough to suggest any. -- Bob Bagwill ------------------------------------------------------------------------------- From: "Angela K. Demke" > > Are you using PVM today? No. (This is not due to any deficiency with PVM, and I expect to be using it again in the next several months. > What are your application(s)? Primarily a parallel rendering application. > How many and what kind of machines have you used with PVM? Up to 25 Sun Sparcstations at a time. I have had no opportunities to take advantages of PVM's heterogenous capabilities. ------------------------------------------------------------------------------- From: Are you using PVM today? Yes What are your application(s)? Determining the electrostatic potential energy of interaction between charged molecules. How many and what kind of machines have you used with PVM? I have used up to 12 DEC Alphas at one time. I am planning to use a heterogeneous environment of Suns and Decs soon, though. Any comments and suggestions for improvement are welcome. Please email the information to pvm-survey@cs.utk.edu. Usuable interrupt routines for the arrival of a message would be great. ------------------------------------------------------------------------------- From: Paul Gray In response to your recent survey request, here is my status on PVM. I have never used PVM, _but_, I am learning PVM with specific interest in applying it to modelling of layered superconductors. Specifically, I'm interested in running PVM on 6 DEC alphas to drive a FORTRAN program. Hopefully, by the time the next survey rolls out, I'll be knee-deep in PVM! -Paul ------------------------------------------------------------------------------- From: acrobin@sandia.gov (Allen C. Robinson) > >Are you using PVM today? I have PVM as a compile time option but it is not used extensively. Mostly it is used to show feasibility. MPI may take precedence in the future. > >What are your application(s)? Shock wave physics > >How many and what kind of machines have you used with PVM? SUN,HP,IBM,SGI - Maximum number of nodes ever = 24 on an SP2. ------------------------------------------------------------------------------- From: Len Myers 1. Yes (am using it now) 2. To support distributed expert systems. (usage) 3. HP 700s, IBM RS6000s, and SUN3s & 4s. (on what machines) -len myers- ------------------------------------------------------------------------------- From: roscoe@BUCRF15.BU.EDU (Roscoe C. Giles) > >Are you using PVM today? Yes, we are. > >What are your application(s)? We (the Center for Computational Science at Boston University) perform a range of scientific and educational computing functions. My particular use for PVM ha been in educating students about parallel computing paradigms and algorithms. > >How many and what kind of machines have you used with PVM? CM-5, SGI, SUN > ============================================================== Roscoe C. Giles Center for Computational Science, and Dept. of Electrical, Computer, and Systems Engineering Boston University 44 Cummington Street, Boston, MA, 02215 (617) 353-6082 roscoe@roscoe.bu.edu http://roscoe.bu.edu/ ------------------------------------------------------------------------------- From: adilger@enel.ucalgary.ca (Andreas Dilger) > Are you using PVM today? Although I have downloaded PVM, I have not had a chance to use it yet. > What are your application(s)? The initial reason for downloading PVM was for running distributed ray-tracing, namely POV-ray. However, there is also a desire to have a distributed MPEG encoding and decoding system. I also want to implement a distributed fractal video compressor as part of my master's thesis. > How many and what kind of machines have you used with PVM? The majority of machines will be SUN SPARC 2, SPARC 5, and SPARC 10 uniprocessor systems. Depending on avaliability, there may also be a large number of DEC 125? systems (with MIPS R3000 processors). > Any comments and suggestions for improvement are welcome. The one feature that I would like to see implemented would be the use of the 'speed' field when assigning hosts. The actual values used would be a metric dependant on the application being run, but if there were some benchmark values given for common architectures, it would avoid the need for storing separate initialization files. The application could then scale its values to give approximately the standard value. Ideally, one should be able to modify this value while running to take into account load conditions, and possibly network transmission time. The load and transmission time should probably be kept as separate values, since a fast machine at a distant location should actually get larger chunks to process, to reduce the transmission lag. One other thing that comes to mind is if it is possibleto distribute a 'C' only library, since nobody uses the Fortran library here, and I have spent some time taking out the fortran parts of the code. My apologies if any of this has already been implemented, since I haven't used the library yet. This is only based on a preliminary look. Otherwise, it looks very useful, and it will save me lots of effort in making distributed code. Have a happy new year. -- Andreas Dilger | University of Calgary "If a man ate a pound of pasta Lab:(403)220-8792 | Micronet Research Group and a pound of antipasto, would Office:(403)220-4716 | Elec & Comp Engineering they cancel out, leaving him still hungry?" -- Dogbert ------------------------------------------------------------------------------- From: arbogast@masc29.rice.edu (Todd Arbogast) We have incorporated a PVM option in our code: RPARSim (Rice Parallel Aquifer and Reservoir Simulator). This simulated petroleum or groundwater subsurface flow, and the transport of dissolved chemical species that may be chemically reactive and/or radioactive. I have run it on a collection of SUN SPARC2's. We call PVM from an "MPI-like" library of generic message passing calls. This library can call the actual message passing calls of Intel, PICL, PVM, CM5, and null (serial). -- Todd Arbogast Department of Computational and Applied Mathematics Rice University ------------------------------------------------------------------------------- From: agaqb@vis-crimson.inre.asu.edu (Ali Benmalek) Here are answers to your questions in your PVM survey: Q: Are you using PVM today? A: Yes, I am using PVM today Q: What are your application(s)? A: Large three-dimensional radiative-transfer finite-element Fortran code. Q: How many and what kind of machines have you used with PVM? A: Used PVM to run the code in parallel in a cluster of workstations. Workstations include multiprocessor SUN SPARC, SGI's (4D-series, Crimson) and IBM RS/6000. Additional comments and suggestions: 1- Need more types of data in PVMFPACK, ect... Important data types for scientific (particularly, engineering and numerical analysis) Fortran applications include DOUBLE-PRECISION real and integer numbers, COMPLEX numbers, and LOGICAL and CHARACTER variables. 2- Need timing subroutines to be used in Fortran and C programs. Now, there are difficulties in evaluating accurately run times of applications in different processors or workstations. 3- Put more features in XPVM. ------------------------------------------------------------------------------- From: klie@masc47.rice.edu (Hector Manuel Klie) > Are you using PVM today? Yes. > > What are your application(s)? Reservoir Simulation Engineering. > > How many and what kind of machines have you used with PVM? > No more than 4, just for debugging purposes... ------------------------------------------------------------------------------- From: shawn@targa.llnl.gov (Shawn Dawson) Are you using PVM today? YES What are your application(s)? ATMOSHPERHIC and LAND modeling simulation at Lawrence Livermore National Laboratory How many and what kind of machines have you used with PVM? 8 SGI workstations ------------------------------------------------------------------------------- From: "Anthony P. Leclerc" Currently, I'm not using PVM. However, last semester, I used PVM in my Parallel Processing class. We used it on a distributed network of Ultrix Workstations. My original plan was to use the Intel Paragon at the University of South Carolina, but this fell through. The applications I run are simple labs (Simpson's rule for integration for example). However, I have written a parallel global optimization solver. I wrote this to work on a distributed network of UNIX workstations using sockets. I will be re-writing using PVM, and hopefully implement it on the Intel Paragon. I enjoyed using PVM. A question for you: If I use PVM for the parallel communication part of my parallel algorithm, can I market it, or do I owe you something, or can I even do this. If not, I will be "cleaning up" my socket code. I hate other people having there hands in what I write. Understandably, however, I see that you may want a share or whatever. Thanks...Tony Leclerc ------------------------------------------------------------------------------- From: yih@atom.cs.utah.edu (Benny Yih) > Are you using PVM today? Only occasionally as a baseline to compare other "lighter" weight protcol implementations. > What are your application(s)? None. I am just running other people's applications [from parkbench, etc.] as benchmarks, since we are experimenting w/ a sender-based protocol as the workstation cluster transport. > How many and what kind of machines have you used with PVM? Mostly HP 9000/700s ------------------------------------------------------------------------------- From: "Alan G. Yoder" > Are you using PVM today? No. > What are your application(s)? Distributed computing research applications. > How many and what kind of machines have you used with PVM? Up to 8, Sparcs and RS/6000's. > Any comments and suggestions for improvement are welcome. Around here, people were waiting for the SP/1 version (using the switch) to come out in version 3.x. I have sort of lost touch with how that is going. The main thing I found frustrating was the difficulty of running applications from arbitrary places; PVM was extremely balky about running things from anywhere except a pathname that was hardwired into the thing at compile time. Cheers, Alan --------------------------------------------------------------------------- Alan G. Yoder agy@cse.nd.edu Dept. of Computer Science and Engineering http://www.nd.edu/~ayoder 384 Fitzpatrick +1 219 631 5772 or 5273 Notre Dame, IN 46556 +1 219 631 9260 dept. fax --------------------------------------------------------------------------- ------------------------------------------------------------------------------- From: chaganti@xanth.msfc.nasa.gov (Kris Changanti-ES84/UAH) Are you using PVM today? Yes What are your application(s)? Class projects for Parallel Processing. How many and what kind of machines have you used with PVM? SGI,SUN,nCUBE,CRAY Thanks Krishna Chaganti ------------------------------------------------------------------------------- From: daasch@ee.pdx.edu (W. Robert Daasch) >Are you using PVM today? > Yes, taking today to mean currently. I didn't run PVM today. >What are your application(s)? > Public domain packages such as Conjugent Gradient. Homebrew software for a project in analog circuit design. Last quarter in a graduate level course in Computational Science. Students were required to use PVM in a small conductor modeling project. >How many and what kind of machines have you used with PVM? > Sun, SPARCs and recently began a move to Linux, X86 ------------------------------------------------------------------------------- From: "Fred M. Staudaher" 1. No, but are developing code locally, which we will conver to PVM, when ported. 2. Application is generation of simulated radar clutter and jamming signals. Large data cube. 3. Haven't used PVM yet, but plan to use MHPCC's SP-2, 50 to 400 nodes. ------------------------------------- Name: Fred M. Staudaher E-mail: fred@airborne.nrl.navy.mil (Fred M. Staudaher) ------------------------------------------------------------------------------- From: avtp@trident.datasys.swri.edu (AVTP) >Are you using PVM today? Yes. >What are your application(s)? We are using PVM to develop distributed/parallel algorithms to calculate visualization forms (i.e., isosurface, countour map, streamlines, etc.). >How many and what kind of machines have you used with PVM? Network of Sun Sparcstations (~8 machines), SparcCenter (6 nodes). >Any comments and suggestions for improvement are welcome. A previous engineer on the project reported trouble in implementing the shared-memory version of PVM on our SparcCenter. I do not know if this was a PVM problem, or if it has been fixed by later versions of PVM. PVM has been very helpful in our development! We hope it will continue to be as we move to multicomputer implementations of our algorithms. One suggestion: it would be useful to have timing or load balancing functions to include in applications. I recently went through an analysis of an algorithm we were developing where I had to insert UNIX timing calls and print them to a file, then develop my own load balancing heuristics. Since these things are commonly done in a parallel environment, it might be useful to include them as a toolkit to PVM. Functions that will "measure" network latency or processor speed would make it easier to *efficiently* port applications to environments with varying performance characteristics. A programmer could include these functions to "measure" the environment when an application begins, then make load distribution decisions using these measurements. This seems like a common enough problem. >Please email the information to pvm-survey@cs.utk.edu. >Your help is much appreciated. ---------------------------------------------------------------------------- Joey Baumgartner baumgart@trident.datasys.swri.edu Computer Integrated Manufacturing (210)522-2494 - Office Automation and Data Systems Division (210)522-5499 - Fax Southwest Research Institute ---------------------------------------------------------------------------- ------------------------------------------------------------------------------- From: Richard Boes 1. Yes, we use PVM. 2. All of are applications are developed inhouse. Most are simulations of naval encounters and tools needed for analysis. PVM is used for messaging between cpu's running different platforms. 3. Our environment is made up of 44 MIPS cpus (SGI's of various sizes) and 10 HP cpus (HP 735) connected with FDDI and switched ethernet. I will forward this message to our programming group for comments and suggestions. -Richard Boes rcb@perseus.nl.nuwc.navy.mil ------------------------------------------------------------------------------- From: "Robert Indik" Are you using PVM today? yes What are your application(s)? Modeling semiconductor light amplifiers+lasers, study of spontaneous pattern formation in PDE's How many and what kind of machines have you used with PVM? CM5 (128 nodes), SP2(256 nodes), Silicon Graphis Indy (1), Crimson(1), Sun spracstation 2, and IPC's (23) ------------------------------------------------------------------------------- From: liy@osram.sylvania.com (Yan-Ming Li) My answers are (1) No, (2) pvm 3.23 installed on hp735 running hpux 9.01, plans to run on 4-5 hp720-735-755. (3) numerical PDE based on method of lines and DAE. Cheers, Yan-ming Li OSRAM SYLVANIA INC. ------------------------------------------------------------------------------- From: Michele Freed > >Are you using PVM today? No but will in late Spring, or maybe MPICH > >What are your application(s)? adaptive routing simulations > >How many and what kind of machines have you used with PVM? etherneted network of SUNS. > >Any comments and suggestions for improvement are welcome. >Please email the information to pvm-survey@cs.utk.edu. MPI allows one to create datastructures to send and has more diverse broadcasting ( according to standard). I really like the ability to send my own data structures over going through the data bytes and re-storing into a structure. Granted the creation of datastructures in MPI is not that straight forward. -Michele Freed email: mfreed@uiuc.edu | voice: 217-244-8409 ------------------------------------------------------------------------------- From: > > Are you using PVM today? No. > What are your application(s)? I wrote a ray tracer using PVM about 9 months ago, in C++ with InterViews. > How many and what kind of machines have you used with PVM? Mostly a couple of dozen Suns, but I tried it out on a MIPS-OS variant with four CPUs. I couldn't get the interface between the console on a Sun and the pvmd on the MIPS box to work though, so I just used Suns. > Any comments and suggestions for improvement are welcome. > Please email the information to pvm-survey@cs.utk.edu. I found PVM easy to use and very useful. I would like to use it further some day. I would even find it interesting to be a part of the PVM project team if the opportunity arose. -- "The climate of Bombay is such that its inhabitants have to live elsewhere." Eric D. Hendrickson Damnit! I can't stop the heterocyclic declination! Central Computing Services University of Minnesota, Twin Cities, USA I came to confess. *I* was the second gunman on the grassy knoll.... Eric.Hendrickson-1@umn.edu 612/626-7761 ------------------------------------------------------------------------------- From: Are you using PVM today? yes What are your application(s)? fluid dynamics proprietary benchmarks. How many and what kind of machines have you used with PVM? Digital 5000 (mips based), 2100, 3000, 7000 8000 (Alpha based) series Any comments and suggestions for improvement are welcome. We are having some difficulty in moving from pvm 3.3 to 3.3.4 and pvm 3.5. I saw Al at SC95 but never got data back to him on what the problem is. I will send you more info as I try to work out what the problem is, but basically I get a "cannot open pvmd" error when I try to add hosts. /Charlie. ------------------------------------------------------------------------------- From: satyan@au-bon-pain.lcs.mit.edu (Satyan R. Coorg) Are you using PVM today? We used PVM for an assignment in a course taught here at MIT by Prof. Arvind ("Dataflow Languages and Computer Architectures"). We primarlily used the PVM to show the contrast between explicit forking/message passing versus implicit parallelism. What are your application(s)? The students programmed a successive-overrelaxation kernel. How many and what kind of machines have you used with PVM? We used it on Sun-4's ------------------------------------------------------------------------------- From: ess@arsc.edu (Michael Ess) Are you using PVM today? Yes What are your application(s)? Molecular modeling and game searches How many and what kind of machines have you used with PVM? Currently on the T3D and a network of SGIs In the past on a network of SUNs and HPs Mike Ess Arctic Region Supercomputing Center Work 907-474-5405 P.O. Box 756020 Fax 907-474-5494 Fairbanks, Ak 99775-6020 ess@arsc.edu ------------------------------------------------------------------------------- From: "Dhaval N. Shah" > > Are you using PVM today? Yes > > What are your application(s)? Running tasks in parallel on a Network of Workstations > > How many and what kind of machines have you used with PVM? SUN workstations, 4 to 6 > > Any comments and suggestions for improvement are welcome. > Please email the information to pvm-survey@cs.utk.edu. Low latency ------------------------------------------------------------------------------- From: tigger@petroglyph.cl.msu.edu (Eric Kasten) > > Are you using PVM today? Yes > > What are your application(s)? Mostly research oriented applications, as well as providing a test bed for systems level programs that may be moved to a kernel level once the algorithm has proven to be useful. These applications include networking and distributed computing across a wide range of topics. > > How many and what kind of machines have you used with PVM? Quantity varies. Anywhere from a few to 50. Suns, Alphas, i386-i586, RS/6000. OSes include Solaris, OSF/1, Linux and AIX. ------------------------------------------------------------------------------- From: sy05@PL122c.EECSCC.Lehigh.EDU (Yue Su) I was using PVM for doing Molecular Dynamics on RISC6000 machines. It would be more helpful if some one can add more sample subroutines specially on the subject of scientific calculations. Thanks. ------------------------------------------------------------------------------- From: > > Are you using PVM today? Yes. > What are your application(s)? Parallel Artificial Intelligence- Implementation of a parallel reasoner (specifically, a description classifier) for Loom, a state-of-the-art knowledge representation system. > How many and what kind of machines have you used with PVM? about 20 workstations, Sparc10s, HP715s, HP720s, Hp730s, and Sun4s. ------------------------------------------------------------------------------- From: Murali.Damodaran@ntu.ac.sg > > Are you using PVM today? > . Yes and just beginning. > What are your application(s)? CFD applications. Explorations in Linear Algebra and Mathematics. Rendering in Computer Graphics applications. > > How many and what kind of machines have you used with PVM? Use 4 SGI Indigos plus other platforms. Plan to include SP2 to the cluster of workstations. =o=o=o=o=o=o=o=o=o=o=o=o=o=o=o=o=o=o=o=o=o=o=o=o=o=o=o=o=o=o=o=o=o=o=o=o=o= Murali Damodaran, Sch. of MPE <> Internet: mmurali@ntuix.ntu.ac.sg Nanyang Technological University <> mdamodaran@ntuvax.ntu.ac.sg Nanyang Avenue, SINGAPORE 2263 <> Phone : 65-799-5599 Fax: 65-791-1859 =o=o=o=o=o=o=o=o=o=o=o=o=o=o=o=o=o=o=o=o=o=o=o=o=o=o=o=o=o=o=o=o=o=o=o=o=o= ------------------------------------------------------------------------------- From: "David Robertson [ITG]" >Are you using PVM today? No. >What are your application(s)? Distributed visualization, for example isosurface generation and rendering. >How many and what kind of machines have you used with PVM? Sun's, Cray's, Maspar's (with front ends), DEC Alpha's, SGI Challenger ------------------------------------------------------------------------------- From: Lindsay Bradford > Are you using PVM today? Yes > What are your application(s)? 1) Finding minimal multiplier solutions to the Greatest Common Divisor of 10 numbers. 2) Once appropriately familiarised with PVM, will attempt to find minimal solution to a problem set by my supervisor. AS yet, I do not know the nature of the problem. > How many and what kind of machines have you used with PVM? 7 Machines at the moment. All are Sparc Stations. > Any comments and suggestions for improvement are welcome. Greater explanation of the Message passing system would be lovely. Detailed examples (and accompanying explanations) on passing messages from several tasks would be of great help. Also.. your discussion on the Resource mamager is minimal. I would like to know exactly what to expect from the Virtual Machine using the Resource Manager. Eg - What is a acceptable system load on a machine before the RM decides no to schedule a task to it? Is the task somehow made 'nice' for machines with small enough loads? The Planet Erbierso is very famous for | My views are MINE! | it's Exotic Fish Meat but is beset by | You can't have 'em! | Deadly Edible Arts Graduates. | | Galaxy #4 - Elite. | linds@cs.uq.oz.au | ------------------------------------------------------------------------------- From: sylvain@cfassp45.harvard.edu (Sylvain G. Korzennik) > > Are you using PVM today? Yes. I have recently upgraded from 3.2.0 to 3.3.6 > What are your application(s)? HPCC applications in Helioseismology, ie specific data analysis > How many and what kind of machines have you used with PVM? Sparc cluster: SS2 and SS10, as development platform Sparc SMP: SS1000 4-8 procs, as production SGI SMP: Power serie (R4400 and R8000), 4-8 procs, as production Intel's Paragon (CCSF, CIT), under developemnt TMC's CM5 (NRL and UMIACS), under developemnt T3D (JPL), under developemnt > Any comments and suggestions for improvement are welcome. PGON: stderr and stdout of children was not yet (at the time) redirected to /tmp/pvml. T3D: pvm_spawn() was not yet (at the time) supported XPVM (under 3.3.6) is a great tool, but it is taking a lot of resources (tk/tcl overhead?). Also, when building `space-time' and `utilization' views from a trace file, it is surprising slow (on a SS10), (un-)zooming features are also slow. I haven't yet found any documentation on pvmgs, that I suspect is the tool that generates the trace file. It would be nice to be able to generate a trace file w/out starting XPVM, and later analyze w/ XPVM that trace file. Cheers, Sylvain. -- http://cfassp45.harvard.edu:1080/ "The world couldn't exist without madmen" Maimonides ------------------------------------------------------------------------------- From: evank@zeus.di.uoa.ariadne-t.gr (Evangelos Koutsompinas) > > Are you using PVM today? > My first involvement with PVM was last year at the QUEEN MARY & WESTFIELD COLLEGE of LONDON UNIVERSITY, England. Over there, I installed PVM and used it as the tool to implement a dissertation project. This year I installed PVM at the network at ATHENS UNIVERSITY, DEPARTMENT OF INFORMATICS, Greece and did a couple of lectures about it. I don't know about QMW COLLEGE, but here in Athens people are going to be using it a lot within the following months. > What are your application(s)? > At QMW I implemented a theoretical model called SAPS (Self Adaptive Parallel Servers) intended for parallel MPI hardware. The outcome is a software platform that provides two types of user interfaces and a load balancing scheme: i) there is a C API for programmers who wish to develop services that can parallelize requests and achive speedup by using worker workstations. The SAPS system automates request decomposition and assignment of work fragments to workers (this also includes data send & receive). A service program must comply to the SPMD model as it is an application spawn over a number of workstations. ii) there is a C API for programmers who wish to use already existent services by providing a transparent way to find the desired service (by using a name service technique) and also transmit the input and collect and patch together the output data. iii) because resourses are not limitless, SAPS is using a given number of workstations (pool) and helps already running services share it. It is using a load balancing real time aloocation of workers from the pool to services always maintaining a uniform workers / task load ratio for all services. At ATHENS UNIVERISTY we are using PVM in order to implement another model using MPI that can be a system independent framework for parallel application development (i.e. this has already been finished for a Parcytec GCel machine). Several students will also take part in this projects by programming modules of the code as dissertation projects. > How many and what kind of machines have you used with PVM? > At QMW I've used a network of Apple MACs running A/UX, Apple's version of UNIX. Eventhough I played with the PVM source code, I couldn't get it to communicate with machines other than MACs as there was a problem with XDR. After that, I used 8 SUN4 and 2 SGI machines together facing no problems. At ATHENS UNIVERSITY we have PVM installed and running on a network of SUN4 (25) and a Convex machine connected with 4 powerfull HP workstations via FDDI. ------------------------------------------------------------------------------- From: kmlee@sun.hallym.ac.kr (Kwang-mo Lee) > Are you using PVM today? Yes, we installed PVM about Sep. 1994. > What are your application(s)? Now, we are testing the PVM systems, and we will use this systems on parallel programming research area. We have a plan to use it on the specific applications such as weather forecasiong, parallel simulation etc. > How many and what kind of machines have you used with PVM? We installed this systems on 5 SUN-4, and 1 SUN-3 systems now but we will extend to some more machines later. Kwang-Mo Lee ========================================================================== Prof. Dr. Kwang-Mo Lee Department of Computer and Information Engineering Hallym University #1 Ockchun-dong Chunchon Kangwon-do 200-702 South Korea Tel +82-361-58-1452, Fax +82-361-56-3421 E-mail : kmlee@sun.hallym.ac.kr ========================================================================== ------------------------------------------------------------------------------- From: Shuichi Terauchi Dear pvm-survey, My name is Shuichi Tearuchi. > Are you using PVM today? Yes. > What are your application(s)? I used "ScaLAPACK". > How many and what kind of machines have you used with PVM? NEC SX-3 and NEC EWS4800. Yours sincerely ------------------------------------------------------------------------ Shuichi Terauchi NEC Corp. 1st Language Development Division e-mail tera@lang1.bs1.fc.nec.co.jp Tel +81-423-33-1256 FAX +81-423-33-1962 ------------------------------------------------------------------------ ------------------------------------------------------------------------------- From: Ted Grzesik > > Are you using PVM today? Yes, but only for testing purposes. > > What are your application(s)? Small, test applications used to test concepts of parallelism. > > How many and what kind of machines have you used with PVM? Network of HP-UX workstations. Using about 4 machines at any one time. > > Any comments and suggestions for improvement are welcome. > Please email the information to pvm-survey@cs.utk.edu. Get rid of xdr encoding in favor of ndr encoding. Ted Grzesik Massachusetts Language Lab Hewlett-Packard Company tedg@apollo.hp.com Chelmsford, MA (508) 436-5959 "Contrariwise", continued Tweedledee, "If it was so, it might be; and if it were so, it would be; but as it isn't, it ain't. That's logic." -- Lewis Carroll ------------------------------------------------------------------------------- From: rkg@iitk.ernet.in (r k ghosh) > Are you using PVM today? > Yes I have just been able to install pvm 3.3.6 today > What are your application(s)? I will use it for teaching a course on Parallel Computing > How many and what kind of machines have you used with PVM? We have just been able to use 10 SUN3 m/c. We would like to see if DEC-Alpha and HP9000 m/c can also be used as additional hosts too alongwith SUN3s. /==================================================================\ | R. K. Ghosh | Tel: (91)(512) 257645/257638 (Off) | | Email: rkg@iitk.ernet.in | (91)(512) 258480 (Res.) | | | Fax: (91)(512)250260/250007 | |------------------------------------------------------------------| | Address Correspondence to: | | Dr. R. K. Ghosh, Department of Computer Science & Engg. | | Indian Institute of Technology, Kanpur 208 016, (INDIA) | \==================================================================/ ______________________________________________________ Gautam Barua, Head CSE & CC, IIT Kanpur 208016 India gb@iitk.ernet.in FAX:+91-512-250260/250007 phone: 257252(O), 258228(R) ------------------------------------------------------------------------------- From: cheekong@iss.nus.sg (Chui Chee Kong) > Are you using PVM today? Yes, I did. > What are your application(s) ? I am building a distributed numerical computing library. > How many and what kind of machines have you used with PVM? in the range of 4 to 16 consists of SUN (SS2, SS10, SS20) and SGI Indigo workstations. more effort should be spent in improving the performance of PVM on high speed network such as ATM. I heard of the PVM-ATM. But I have not tried it out. hope the info is sufficient. chee kong ------------------------------------------------------------------------------- From: Aad van der Steen Dear PVM Team, Answer 1: Yes, we are using PVM3 currently. Answer 2: Presently we use PVM for the solution of sparse linear systems and for many dense linear algebra operations. Answer 3: We use PVM on SUNs (SunOs 4, Solaris), DEC Alpha, Convex C220, and SGI R4400 systems. Best Regards, Aad van der Steen ------------------------------------------------------------------------------- From: cwh@ltthp5.epfl.ch (Carl HUSTAD) > > Are you using PVM today? Yes > > What are your application(s)? Unstructured parallel Navier-Stokes computations for 3D turbomachinery CFD. > > How many and what kind of machines have you used with PVM? On cluster of RISC based workstations ( 8 HP-730's) and Cray T3D with upto 256 processors. > > Any comments and suggestions for improvement are welcome. PVM is slow on the T3D, hence I use the Cray shmem routines. This doubles performance. Again on workstation clusters I find that best size for my type of problem is 3 to 6 machines. > Please email the information to pvm-survey@cs.utk.edu. > I think PVM is an excellent tool and is an essential first step for industry wanting to start in parallel computing. The reason for this is because the software is public domain and runs on their existing hardware. Keep up the good work. Regards Carl Hustad, LTT/EPFL, Switzerland. > ------------------------------------------------------------------------------- From: I used PVM for the last time a few months ago. The application is CFD. At the moment this is 3D Euler, but the intention for the futere is 3D Navier-Stokes. The reason why I am not using it today it that I am developing new schemes and this takes a while. I use the Digital ALPHA and I have done some test problems with 4 machines. Edwin. ------------------------------------------------------------------------------- From: stevenso@win.tue.nl (Rob Stevenson) > > Are you using PVM today? I used PVM about a year ago. > > What are your application(s)? My application was multi-grid. > > How many and what kind of machines have you used with PVM? About twenty SUN workstations. ------------------------------------------------------------------------------- From: Markus Schwehm > Are you using PVM today? Yes > What are your application(s)? My application is a distributed genetic algorithm package (VPGA). The package had been developed last year and is now used to solve a large scale manufacturing problem (optimal partition and kanban allocation for JIT production lines). > How many and what kind of machines have you used with PVM? The package has been tested on a (homogeneous) cluster of 23 SUN workstations and on a smaller cluster of Linux PCs. We were not able to compile it on HP workstations (exept for a single-workstation version) > Any comments and suggestions for improvement are welcome. A connection of PVM to the MasPar MP-1/2 woud be appreciated. We use PVM mainly for starting and initializing the master-slave-type algorithm. The actual data communication is realized via files, which is faster as we have measured (only one process change instead of two). Moreover this allows to continue the algorithm after a breakdown. ___________________________________________________________________/_____ Markus Schwehm Tel.: x/9131/85-7617 M assiv Uni Erlangen - IMMD 7 Fax.: x/9131/85-7409 P arallele Martensstr. 3 http://www7.informatik.uni-erlangen.de/~schwehm G enetische D-91058 Erlangen schwehm@immd7.informatik.uni-erlangen.de A lgorithmen ------------------------------------------------------------------------------- From: thorhaue@adam.cs.TU-Magdeburg.DE (Steffen Thorhauer) Dear PVM Project Team, > Are you using PVM today? Yes I use pvm 3.3.5 regularly at home. > What are your application(s)? I use the pvmpovray patch and so I use the parallel povray program. Povray is a raytracer. > How many and what kind of machines have you used with PVM? 4 pc: intel/amd 486/66MHz linux 1.1.59 I'm looking for a good fractal program for pvm with more fractal types than the xep demo program. Thanks. Steffen -- ********************************************************************** * EMail: thorhaue@sunpool.cs.tu-magdeburg.de * ********************************************************************** ------------------------------------------------------------------------------- From: haberj@mathematik.tu-muenchen.de (Joerg Haber) > > Are you using PVM today? Yes. > What are your application(s)? Computer graphics (ray tracing, radiosity). > How many and what kind of machines have you used with PVM? At most 110 HP 9000/720 + 4 SGI Indigo + 5 SUN SparcStation[2,10] + 2 PC (Linux). During development just 2-3 Indigos and 2 PC's. > Any comments and suggestions for improvement are welcome. > Please email the information to pvm-survey@cs.utk.edu. What about a function that gives me the current speed of a machine (depending on current load and processor speed)? Would be helpful for load balancing... Thanks a lot for making PVM available to the public. For me it is very helpful and efficient. Regards, Joerg -- Joerg Haber (SCM) EMail: haberj@mathematik.tu-muenchen.de Mathematisches Institut Tel.: 49-89-21058128 TU Muenchen D-80290 Muenchen ------------------------------------------------------------------------------- From: Zdenek Salvet > Are you using PVM today? Yes. > What are your application(s)? Paralelization of graph algorithms. > How many and what kind of machines have you used with PVM? ~15 machines : SUN4SOL2,SGI,PC/NetBSD -- Zdenek Salvet salvet@informatics.muni.cz ---------------------------------------------------------------------------- If Murphy's Law can go wrong, it will. ------------------------------------------------------------------------------- From: truong@onera.fr (tel 31-87) Dear PVM project team, I'm not using PVM today, but I plan to use it in a near future to compute loads on a helicopter blade. Sincerely yours, Truong V. K. ------------------------------------------------------------------------------- From: Seon-Tae Kim Are you using PVM today? yes, i using a today. What are your application(s)? nummerical calculation. How many and what kind of machines have you used with PVM? sun4 and linux. ------------------------------------------------------------------------------- From: Utz.Wever@zfe.siemens.de (Utz Wever) Dear Colleagues, we are applying PVM very successfully in VLSI, see (Schneider, M., Wever, U. Zheng, Q.: Parallel Harmonic Balance. VLSI Grenoble (1993) ) and (Kleis, U., Wallat, O., Wever, U., Zheng, Q.: Domain Decomposition Methods for Circuit Simulation. Proceedings of for Circuit Simulation. Proceedings of the 8th Workshop on Parallel and Distributed Simulation. Edinburgh (1994)). We are also working on Domain Decomposition methods for drift diffusion equations in process simulation (dopand diffusion or point defect simulation). Most of the time we combine about 10 workstations of the architectures SUN4 and HPPA (mixed). We do not succeed in combining the architectures SUN4 and SUNMP (solaris). The solaris processes do not answer. Is there any bug known ? Finally we want to thank you for your effort in developing this outstanding software. If one want to be portable, there is no alternative to PVM. With best regards U. Wever ------------------------------------------------------------------------------- From: Olivier.Ondoa@imag.fr (Olivier ONDOA) > Could you take a moment to answer three questions. Sure. > Are you using PVM today? Yes, I am actually completing an implementation of PVM over an ATM simulated network, in order to perform some network performance evaluation studies, and to prepare an eventual implementation of PVM over ATM physical network, similar to the version 3.3.2 released by the Distributed Multimedia Center and Computer Science Departement of the University of Minnesota at Minneapolis, in collaboration with both Minnesota Supercomputer Center Inc. and Army High-Performance Computing Research Center. > What are your application(s)? I wrote two popular distributed applications: parallel matrix multiplication and parallel linear equations system resolution using the the Gauss algorithm, to become familiar with the system. But, as soon as I finish the aforementionned implementation, I need some PVM Benchmarks for simulation purposes. Then if any body knows where and how to get these Benchmarks, please let me know. > How many and what kind of machines have you used with PVM? I have been using up to 10 (it could be largely more) workstations, running UNIX and interconnected via an Ethernet LAN. Our workstations set consist of Sparc (5,10,20) and Sun IPX machines. > Any comments and suggestions for improvement are welcome. I am not going to be original by saying that, the PVM system is great tool. > Best wishes, > The PVM Project Team And to you. Regards, Olivier (Olivier.Ondoa@imag.fr). ------------------------------------------------------------------------------- From: chris.eaton@aea.orgn.uk (Christine Eaton 3277) We have done some preliminary studies, and are about to start on a real project using PVM. The application is a Monte Carlo radiation shielding code, called MCBEND. This tracks neutrons and photons through complex geometries. Being a Monte Carlo code it can take a long time to get good statistics. A customer in the oil well logging field - a big application area for MCBEND - has installed an IBM supercomputer, made up of 28 IBM RISC processors. We plan to produce a Parallel MCBEND for them. We use SUN Microsystems workstations, so we will develop the PVM MCBEND using our own machines, of which we have a large number, then port it to the AIX environment. We have found the recent manual and the whole PVM system easy to use. Ewan Webster AEA Technology ------------------------------------------------------------------------------- From: Eric S Fraga pvm-survey@edu.utk.cs writes: > Are you using PVM today? yes > What are your application(s)? optimization in chemical engineering. specifically, for the automated synthesis of chemical processes. also for the development of a multiuser multitasking design environment. > How many and what kind of machines have you used with PVM? up to 13 networked Sun SPARCstations. > Any comments and suggestions for improvement are welcome. none. i'm very happy with PVM 3. -- ........oo...ooo..o..o.o...o.o....oooo..oo...o.....o.....ooo..o.............. Eric S Fraga, ECOSSE group, Dept of Chemical Engineering, Univ of Edinburgh phone: +44 (31) 650 5065/8565, eric.fraga@ed.ac.uk na.fraga@na-net.ornl.gov URL http://www.chemeng.ed.ac.uk/people/eric.html ------------------------------------------------------------------------------- From: Sabina Rips Hallo! > Are you using PVM today? yes > > What are your application(s)? distributed runtime-systems for concurrent logic programming languages > > How many and what kind of machines have you used with PVM? 4 Sun workstations (SunOS 4.1.3) (2 SPARCs, 1 ELC, 1 Classic) Bye, Sabina ------------------------------------------------------------------------------- From: vinicio@ele.puc-rio.br (Enrique Vinicio Erazo [94.1]) PVM survey Dear PVM Project Team, I'm using PVM too. I have two principal groups of applications running on PVM: - A library to implement a distributed system environment that present a high-level interface based in reconfigurable, interconnectable objects. This methodology intend to abstract the moduling, interfacing, configuration, reuse, and paralleling concepts of software systems in order to simplify user's work. A lot of examples were building using this library: computing of PI, solution of Poisson's equation, etc. - A system for artificial vision (image classification) using artificial neural networks. For reduce the training times and the recognition times, the system was paralleling using the former library. The computers used to run these applications are 15 workstations SUN4 (SparcStation 2 with SunOS 4.2) linked with a Ethernet network. Suggestion: Is very important in order that applications with extern configuration can porting your code to any machine with PVM, create PVM primitives for join and leave a process -different to the calling process- to determinate group. Example: int inum = pvm_jointogroup(int tid, char *group) int inum = pvm_lvfromgroup(int tid, char *group) where "tid" is an integer task identifier and "group" is a character string group name of an existing group. The changes in the Group Server code (pvmgs.c) aren't very complex. Enrique Vinicio Carrera E. Catholic University - Rio de Janeiro, Brazil e-mail: vinicio@ele.puc-rio.br vinicio@cygnus.gsc.ele.puc-rio.br ------------------------------------------------------------------------------- From: Kris Van Alsenoy Dear PVM-team, Yes, I am using PVM for all my calculations. The calculations I perform are ab-initio Hartree-Fock calculations on molecules. I use my own program (called BRABO) in which I implemented your PVM-system and use it to perform calculations on large molecules (number of basisfunctions > 1000 (if this has any meaning to you)). The program runs on a cluster of 3 RS6K's (375, 365 + 320H) and separately on a mixed cluster of 3 RS6K's () and 2 DECstations (5000/120). At the moment the DECstations are being replaced by a SUNspark-20 and I hope to be using PVM on this node shortly. Hoping this was of some help, sincerely, kris van alsenoy ------------------------------------------------------------------------------- From: Hans Leichtl ------------------------------ Start of body part 1 Dear PVM-Team, I'm studying computer science at the Technical University in Regensburg, Germany. We uses PVM in one of our classes dealing with parallel computing. We implemented 2 algorithms (soliaire and gauss-seidel) on a SUN pool with Sparc Classic and Sparc 5 stations (15 machines alltogether) to get a feel for the advantages and problems related with parallel computing/programming. So it was a rather short usage of PVM. But I think it's a good tool to use resources of workstation pools and give people an insight on parallel computing. I hope I could help you a little. -hans ------------------------------ Start of body part 2 ------------------------------------------------- hans.leichtl@rz.fh-regensburg.d400.de computer science department fh regensburg; germany ------------------------------------------------- ------------------------------------------------------------------------------- From: lp01 * * Are you using PVM today? yes * * What are your application(s)? computational mechanics codes. development of semi automatic tools for conversion of serial f77 to parallel messages passing f77 e.g. pvm, mpi etc * * How many and what kind of machines have you used with PVM? sun workstations, up to 12, but speed up only on about 4 on average * * Any comments and suggestions for improvement are welcome. * Please email the information to pvm-survey@cs.utk.edu. biggest problem for us is communication latency. Debugging using sun debugger tools has been dificult using pvm 3.x, slaves seem to hang when attaching debuggers and then continuing execution. (this is 3.1.2 under solaris 2.3 using 3.0 compilers/debuggers) -- Peter F. Leggett -- Researcher/Lecturer (PLEASE NOTE NEW PHONE AND FAX NUMBERS, EFFECTIVE IMMEDIATELY) Parallel Processing Group Centre for Numerical Modelling and Process Analysis University of Greenwich Wellington Street Woolwich Tel : +44 (0)181-331-8731 London SE18 6PF Fax : +44 (0)181-331-8665 \,,,/ Email : P.Leggett@gre.ac.uk (o o) --------------------------------------------oOO--(_)--OOo----------------| ------------------------------------------------------------------------------- From: davide@di.unito.it (Davide Cavagnino ) I'm a Ph. D. student in Turin (Italy) and I was interested in PVM as a powerful way to distribute computing. I used the manual only to do a lesson as a Ph. D. exam but I never used your software even if I hope I'll use it in the future. Thank you for giving me the possibility to use the documentation. Davide Cavagnino ------------------------------------------------------------------------------- From: boehm@nolde.informatik.uni-koeln.de (Max Boehm) > Are you using PVM today We plan to use PVM for a portable distributed workload balancing algorithm. > What are your application(s)? Solving of combinatorical optimization problems (SAT, ...). > How many and what kind of machines have you used with PVM? PowerXplorer with currently 8 nodes (based on the PowerPC chip). of Parsytec. We use a homogenous version of PowerPVM 3.2 with the PARIXPPC.H architecture (beta implementation of parsytec). - Max Boehm ------------------------------------------------------------------------------ Tel: +49 221/470-5379 email: boehm@informatik.uni-koeln.de Fax: +49 221/470-5387 Institut fuer Informatik, Pohligstr.1, D-50969 Koeln ------------------------------------------------------------------------------ ------------------------------------------------------------------------------- From: cfred@ipanema.nce.ufrj.br (Carlos Frederico (Adriano)) Answers to the survey: Are you using pvm today? Yes What are your applications? three basic applications: 1> a gaussian elimination program to solve linear equations systems 2> a convolution program (image processing program) for smoothing a greyscale image (represented by a 2D matrix) 3> fractal generation program How many and what ...? a cluster of 8 SUN SPARCstations ------------------------------------------------------------------------------- From: rdb@roke.co.uk (Richard Beton SDH) At Roke Manor Research Ltd (a company owned by Siemens AG of Germany) we are using PVM 3.3.5 for several applications: *) numerical simulations, e.g. radar cross section prediction on a parallel processor consisting of i860 CPUs with transputer communication processors, coupled to external Suns and a DEC Alpha. PVM is particularly used for the workstation interconnection. *) general workstation connectivity, e.g. data exchange between HP and Sun workstations for a multi-workstation medical imaging application. One workstation is dedicated to the MRI scanning image capture, the other to visualisation, analysis, and surgical therapy planning and assistance. *) homogeneous workstation connectivity: parallel 'make' of the software of a larger project. This uses up to 10 Sparcstations for concurrent compilation of separate software modules for a particular project which consist of hundreds of such modules. The workstations on which we have used PVM are: Suns running SunOS4 Suns running SunOS5 HP 700 series running HP-UX Dec Alpha runnning OSF/1 I have one comment regarding the SunOS5 port: PVM 3.3.3 had a particular difficulty in starting up pvmd daemons. Once up, the daemons operate correctly until terminated (always with error). PVM 3.3.5 has improved on this but it is still not perfect. Although it is now much easier to start the daemons, they still fail more often than is convenient. Using the console to add a host, the error message when a failure occurs is libpvm [t40001]: pvm_addhosts(): Pvmd system error with [t80040000] waitc_get() cod dm_startack from t80000000 wid 262149 not found added to the /tmp/pvml. file. I hope this information is useful to you. Regards, Rick. -------------------------------------------------- Richard Beton Roke Manor Research Ltd Romsey, Hampshire SO50 0ZN (01794) 833591 fax (01794) 833433 -------------------------------------------------- ------------------------------------------------------------------------------- From: "Frederico Liporace" I'm not using PVM now, but intend to use it to train neural networks. My cluster configuration consists of 5 IBM RS/6K workstations connected by a FDDI ring. Hope to have more to tell in the next year's survey! best regards, Fred PS. One item of my wish list: a collection of C++ classes that encapsulate PVM. ------------------------------------------------------------------------------- From: dear colleague, At present we are just starting the use of PVM on an eight-node SP2 IBM computer and we have the opportunity to begin the use of T3D Cray computer. Sincerely yours, D. Levesque ------------------------------------------------------------------------------- From: Christian.Kebekus@euler.mathematik.uni-dortmund.de (Christian Kebekus) Subject: Re: PVM survey >Are you using PVM today? Yes. >What are your application(s)? I'm using PVM for a finite elements package to solve partial differential equations (especially I use Domain Decomposition Methods). >How many and what kind of machines have you used with PVM? At this moment I use PVM on four SUN Sparc10 workstation. In the future I want to translate my program to a Parstec Explorer system with 8 nodes (MPS601, 32MByte). _| _| _| _|_|_| 1, rue Maurice Audin _| _|_| _|_| _| _| 69518 Vaulx-en-Velin cedex, France _| _| _|_| _| _| tel (33) 72 04 70 33 _| _| _| _| _| sec (33) 72 04 70 31 _| _| _| _| _| _| fax (33) 72 04 70 41 _|_|_| _| _| _|_|_| @-mail cornet@volnay.entpe.fr ------------------------------------------------------------------------------- From: Dear The PVM Project Team. My name is Eiju Kanazawa.I have two E-mail adress, kanazawa@nsis.cl.nec.co.jp and kanazawa@icfd.co.jp. These adresses are same. I live in Tokyo, I use kanazawa@icfd.co.jp and I live in Tukuba, I use kanazawa@nsis.cl.nec.co.jp. Don't mind my E-mail adresses, and my poor English...... if you send E-mail, plese send to kanazawa@nsis.cl.nec.co.jp. I thank PVM and PVM Project Team for enjoy parallel programing, first. >Are you using PVM today? No. I don't use PVM, in the last 1 month. >What are your application(s)? I made very simple laplace equation, and made animetion. >How many and what kind of machines have you used with PVM? I used NEC EWS4800/210, EWS4800/260, EWS4800/350, PC9801RA21+PANIX(OS) SONY NEWS3870 SGI IRIS4D/210, IRIS4D/350 CRY YMP/EL4 CONVEC C2J/IP IBM RS6K/320H CM? CM2 >Any comments and suggestions for improvement are welcome. >Please email the information to pvm-survey@cs.utk.edu. I am very grad that heteros machines can sends and gets the data. it is very nice.but ,even now , I don't use and understand dynamic load average system.pvm console command displays same machines ability. same day, I will try PVM. Sincerely yours, Eiju Kanazawa ------------------------------------------------------------------------------- From: Dick Wilmoth > > Are you using PVM today? > Yes. > What are your application(s)? > Direct Simulation Monte Carlo (DSMC) analysis of rarefied gas flows. (Space Shuttle, various orbiting and reentry vehicles) > How many and what kind of machines have you used with PVM? > Currently using: Three 4-processor Sparcstation 10 Two 2-processor " " Four 1-processor " " Have used PVM in past on Cray YMP, C90. ------------------------------------------------------------------------------- From: tannenba@engr.wisc.edu (Todd Tannenbaum) > >Are you using PVM today? > Yes. I am the director of the "Model Advanced Facilities" lab in the College of Engineering at Univ of Wisconsin-Madison College of Engineering. Folks who use the lab rely **heavily** upon PVM. PVM is used to develop/test parallel applications on our cluster of FDDI-connected HP and Alpha workstations. Several folks then migrate to an IBM SP2 (using PVMe) for production runs. >What are your application(s)? > Engineering modeling & simulation; lots in the area of Chemical Engineering. Also work with sparse matricies. >How many and what kind of machines have you used with PVM? > Hewlett Packard 735 Workstations DEC Alpha Workstaions Sun SPARCs IBM RS/6000s IBM SP1 and SP2 >Any comments and suggestions for improvement are welcome. Better support for batch-style scheduling and processing. Ideally, better integration with the CONDOR distributed-batch/scheduling system would be ideal. PVM and Condor currently work together, but in a somewhat limited fasion. A packet "redirection" mechanism which would allow a scheduler to move a compute node from one CPU to another, etc. An API interface to PVMD which would be used not by end-user programs, but by scheduling software. The API could provide a whole host of functions that schedulers would love, but would be difficult to accomplish "outside" of pvmd proper. i.e. things to gracfully suspend, kill, migrate, etc, a compute node. Better disaster recovery and cleanup. By cleanup, I mean if things blowup then pvmd does not hang, removes lock files, etc, from /tmp, etc. P.S.: a heartfelt "Thank You" to the PVM Project Team for all their hard & excellent work! Esp in Chemical Engineering polymer studes, here at Univ of Wisconsin MAF we are making great new strides & discoveries based upon simulation code written in PVM. =-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-= Todd Tannenbaum email: tannenba@engr.wisc.edu Director of the Model Advanced Facility voice phone: (608) 262-3118 Computer-Aided Engineering Center FAX phone: (608) 265-4546 University of Wisconsin-Madison ------------------------------------------------------------------------------- From: T.G.Thomas@qmw.ac.uk (T.G.Thomas) > Are you using PVM today? Yes. > What are your application(s)? Computational Fluid Dynamics (Turbulence simulation using Large Eddy Simulation / Direct Simulation techniques) > How many and what kind of machines have you used with PVM? Code development on a few networked SGI Indies and Linux PCs. Production work on Cray T3D and Intel iPSC/860. -- Glyn Thomas | Turbulence Unit, Dept Civil Engng, Internet: t.g.thomas@qmw.ac.uk | Queen Mary & Westfield College, Voice: +44 (0)71-975-5196 | Mile End Rd, London E1 4NS, UK. ------------------------------------------------------------------------------- From: Anthony B Wilkinson I use it in a parallel programming class every year, with SUN workstations (usually just a few). We did have to write our own software to get around not having rlogin/rsh turned on on our network and introduce local security in the past. It would be nice to have a meeting/workshop on using pvm for teaching. Barry Willkinson Assoc. Professor Dept of Computer Science UNC-Charlotte ------------------------------------------------------------------------------- From: kitajima@cic.unb.br (Joao Paulo Kitajima) > Could you take a moment to answer three questions. Just a remark: I do not work anymore in France. I am in Brazil now. My new address is in the signature of this message. > Are you using PVM today? Yes. > What are your application(s)? I have developed a tool that simulates (using synthetic workloads) a parallel program. Program written in C. > How many and what kind of machines have you used with PVM? 32 RS6000 inside a SP-1 parallel machine. > Any comments and suggestions for improvement are welcome. > Please email the information to pvm-survey@cs.utk.edu. For me currently, PVM is good as it is today. Joao Paulo -- Kitajima, Joao Paulo (CNPq fellowship) (kitajima@cic.unb.br) CIC (Ciencia da Computacao) vox +55 (0)61 348 2482/fax +55 (0)61 273 3589 UnB - Campus Universitario, Asa Norte 70919-900 Brasilia DF - Brasil ------------------------------------------------------------------------------- From: Matthijs Kadijk > Could you take a moment to answer three questions. > > Are you using PVM today? YES. > > What are your application(s)? (Nonlinear) Finite Element Calculations. Currenlty I am devellopping a distributed database management system for a special purpose database containing a large FE model. > How many and what kind of machines have you used with PVM? not more that about 5 machines per VM from LAN with HP, SGI, SUN (SPARC), DEC ALPHA (and Linux) That's all folks, Matthijs. _____________________ ______________________________ / Matthijs Kadijk \ / email: kkm@bouw.tno.nl \ | TNO-Bouw, Postbus 49 | www: http://www.bouw.tno.nl \___________________ | NL-2600 AA Delft | tel: +31 - 15 - 842 195 /\ fax: +31 15 843975 \ \_____________________/ \ ________________________/ \_ _____________________/ ------------------------------------------------------------------------------- From: rsh@ddci.dk (Rene Staune Holm) >Are you using PVM today? no >What are your application(s)? A Fortran77 program. Solving Navier Stokes equations. >Machines used with PVM A SPARCstation2 and three SPARC IPX running SunOS 4.1.3. Regards, Rene S. Holm (rsh@ddci.dk) DDC-I A/S ------------------------------------------------------------------------------- From: Peter-Christian.Eccardt@zfe.siemens.de (Peter-Christian Eccardt) > >Are you using PVM today? > yes >What are your application(s)? > we try to couple simulators running on different workstations within a noncommercial project supported by the german government >How many and what kind of machines have you used with PVM? > 3 SGI (4D25, indigo, indy) 1 SUN Sparc II >Any comments and suggestions for improvement are welcome. >Please email the information to pvm-survey@cs.utk.edu. > PVM was very helpful for us. As we are developing our software with c++, we wrote a c++ class library based on PVM to make pvm even more easy to use as it already is. Please let us know about new developments Best regards -- Peter-Christian Eccardt Siemens AG ZFE T KM 1 Otto-Hahn-Ring 6 81730 Muenchen Tel. (089) 636-47689 Fax (089) 636-46881 email eccardt@ztivax.zfe.siemens.de ------------------------------------------------------------------------------- From: eva@ac.upc.es (Eva Lopez Cervilla) >Are you using PVM today? Yes. >What are your application(s)? Some taken from the APPARC Application Suite and parallized by myself. All they solve sparse matrices. We are using them to study the behaviour of parallized applications in message-passing systems. Basically, we are measuring how execution time varies with communication parameters such are data rate of transfer between processors, number of processors...etc. Were you interested on any special feature, just let me know and I'll send you back the information. >How many and what kind of machines have you used with PVM? I've been using it with two different kinds of machines: ALPHA _____ Caractheristics : System name : OSF1 Node name : romeu Operating system release : V2.0 Operating system version : 240 Machine hardware name : DEC Alpha CNVX _____ Caractheristics : System name : Convex O.S. Node name : kefren Operating system release : V10.1 Operating system version : 1 Machine hardware name : Convex C-series and only one machine of each kind. >Any comments and suggestions for improvement are welcome. Well, I've been having a problem with a large application, that passes large messages, when the number of processors running in parallel is also "large" (13). I got messages of the kind: "Task xxxxxxx not responding. Marking task xxxxxx as dead". Noone here seems to know why that happens. My own suspicion is that PVM has an internal counter in order to detect whether a task is dead or not, and, because of the large times spent in communications, some task is unable to update its counter, so PVM doesn't see it any more. Should it be true, could you give PVM users some kind of information about where this counter is located and how it works? Should it not, do you have any idea of why it's passing? (Strictly talking, that shouldn't be my problem. I'm just a user and shouldn't have anything to do with PVM internals. But, if you've got the answers, they will be welcome). Second: In Convex, we have got a system of queues, and interactive CPU time per user is limited. My way of working is as follows: I send the parent application to the queue. There, it starts PVM, and, once started, spawns the rest of tasks that will run in parallel. As I have experienced hundred of times, PVM and children tasks consume interactive CPU time, in spite of having been started from into the queue and not having any terminal assigned. Now, let's suppose that you've got eight tasks running in parallel and that seven of them plus the daemon are consuming interactive CPU time. How long would you be able to run before you get the message: "You've reached your i. CPU quote. All your interactive processes have been killed"? Got the idea? No need to say that Convex experts assure it's a PVM bug. Do you have anything to argue? If you are wishing to work on any of this and need additional information, please, email me. I appreciate your interest. --------- Eva ------------------------------------------------------------------------------- From: nystrom@ludwig.psc.edu (N.A. Nystrom) > Are you using PVM today? Yes. > What are your application(s)? Currently: GAMESS (ab initio quantum chemistry) > How many and what kind of machines have you used with PVM? 6 machine types: Cray T3D, C90 SGI Indy, Crimson DEC Alpha, PMAX > Any comments and suggestions for improvement are welcome. I'd like to see a clearer specification of the implementation of PVM. For instance, the different location of pvmd3 on various systems often causes confusion for new users. Nick Nystrom PSC ============================================================================ Nicholas A. Nystrom, Ph.D. Pittsburgh Supercomputing Center Scientific Specialist Mellon Institute Building email: nystrom@psc.edu 4400 Fifth Avenue voice: +1 412-268-1592 fax: -5832 Pittsburgh, PA 15213 USA ============================================================================ ------------------------------------------------------------------------------- From: Davide Anguita > > Are you using PVM today? yes > > What are your application(s)? > Neural Network Simulation > How many and what kind of machines have you used with PVM? > IBM R6000/550 - IBM POWERPC/250 - HP 9000/720 > Any comments and suggestions for improvement are welcome. > Please email the information to pvm-survey@cs.utk.edu. > -- Davide. ========================================================================== Davide Anguita DIBE Phone: +39-10-3532192 University of Genova Fax: +39-10-3532175 Via all'Opera Pia 11a e-mail: anguita@dibe.unige.it 16145 Genova, ITALY ========================================================================== ------------------------------------------------------------------------------- From: "Ernest L. Baker" (AED-EWD) >>Are you using PVM today? Yes. >>What are your application(s)? Nonlinear optimization. >>How many and what kind of machines have you used with PVM? SGI 4D, 4 CPUs. Cray T3D planned. >>Any comments and suggestions for improvement are welcome. New user. Premature for meaningful comment. >>Your help is much appreciated. Thank you for your help. ------------------------------------------------------------------------------- From: rlh0750@tntech.edu (Roger L. Haggard) > >Are you using PVM today? No. >What are your application(s)? Teaching parallel processing - relatively simple example programs: matrix algebra and ODE/PDE solutions. >How many and what kind of machines have you used with PVM? On a network of DEC workstations at Pittsburgh Supercomputer Center (about 8 to 10 machines). > --------------------------------------------------------------------- Dr. Roger L. Haggard, Assistant Professor EE Department, Box 5004 FAX: (615)372-6172 Tennessee Tech University Email: RLH0750@TNTECH.EDU Cookeville, TN 38505 Phone: (615)372-3453 --------------------------------------------------------------------- ------------------------------------------------------------------------------- From: wwingle@gale.Mines.Colorado.EDU (William Wingle - GE) PVM Project Team, > Are you using PVM today? I've ported PWM over, but I haven't had the time to use it yet. > What are your application(s)? Eventually I will be applying it to geostatistical methods applied to contaminant groundwater issues. Others in my lab apply it to 3D seismic inversion problems (geophysics) for oil and gas reservoir analysis > How many and what kind of machines have you used with PVM? Again, I haven't applied it let, but we have pvm running on 25 IBM RISC 6000 and 10 Next workstations. Bill Wingle Department of Geology and Geological Engineering Colorado School of Mines Golden, Colorado 80401 (303) 273-3905, (303) 273-3859 (FAX) wwingle@mines.colorado.edu ------------------------------------------------------------------------------- From: vspa@eng.auth.gr (Vasilis Spais) Dear Sirs, I am briefly responding to your query concerning our use of PVM: Yes we are using PVM today (unfortunately version 3.2; where can I get the latest version including documentation? I have version 3.6 undocumented) We are using it to run Genetic Algorithms (an embarassingly parallel application) and we are trying to use it for numerically solving electromagnetic field propagation problems by applying a parallel version of the Beam Propagation Method. We run PVM over a slow ethernet network on Unix workstations, mainly series HP9000/700. We usually employ two to four workstations, although we have occasionally used as many as eight. Vasilios A. Spais Electrical Engineer ------------------------------------------------------------------------------- From: andrea@cip.unica.it (Andrea Bosin) Answers to your questions. > > Are you using PVM today? Used up to november 1994; planned to use it again from april 1995. > > What are your application(s)? Quantum Monte Carlo simulations for simple atoms and solids (future). > > How many and what kind of machines have you used with PVM? Usually less than 10, twice or three times around 30 (on a IBM SP1). Sun SPARCstation, IBM RS6000, HP Apollo, Digital Alpha. > > Any comments and suggestions for improvement are welcome. You are working very well! ------------------------------------------------------------------------------- From: "Peter Krull" > > Are you using PVM today? Yes. > > What are your application(s)? We have developed our own software to process satellite imagery. We use PVM to perform simple message passing between distributed hosts. > > How many and what kind of machines have you used with PVM? 4 - 16 single cpu SGI Indigo/Indigo2's and 2 multiple cpu SGI Onyxs. > > Any comments and suggestions for improvement are welcome. > Please email the information to pvm-survey@cs.utk.edu. > It would be nice to be able to have PVM be reentrant (if I'm using the term correctly). I've run in to problems when I've wanted to monitor host utilization. My desire is to send a signal to the host and have it report back usage stats. I've found that if the host is blocking while waiting for a message and gets a signal, it doesn't return from the signal handler successfully. I was told by PVM support that because signals are handled differently on different platforms, a solution to this problem is probably not feasible. My only suggestion for a solution would be to have platform specific versions of PVM. This problem is probably not significant enough to justify such a solution, however. -- Peter Krull Space Dynamics Lab 1747 N Research Parkway Logan, Ut 84341 801 755-4357 pete@samson.idec.sdl.usu.edu ------------------------------------------------------------------------------- From: "Robert G. Brown" > > Are you using PVM today? Yes and no. I'm not using it "today", but I'm using it as part of the project I'm working on (just not that part I'm CURRENTLY working on). > What are your application(s)? Dynamical numericalj simulation of d-lattice-dimension O(n) Heisenberg Models. > How many and what kind of machines have you used with PVM? A farm of up to 32 Sparcstations of varying vintage (single SS-I desktop CPUs through 4x Ross HyperSparc CPUs in SS-10's and 630's). I made some effort to run on heterogeneous machines, but it was too difficult to keep code consistent across platforms that lacked an ANSI C compiler or had different systems libraries. Right now I'm working primarily on a farm of Sparc CPU's that gives me around 250 MFLOPS sustained on coarsely parallelized von Neummann code running in master/slave mode, WHILE other folks are using the consoles of many of those machines for standard Unix activities. The farm is scalable, and I hope to get it up to .5-1 GFLOPS over the next year as we upgrade desktop computing facilities in our department and install additional compute servers. > Any comments and suggestions for improvement are welcome. Hmmm, it has been nine months or more since I last worked with it, but as I recall there are several features that I would like to see implemented or improved. 1) The processor monitor/control facility is (to say the least) primitive. It should return as much information as ps -auxww (BSD) does on the machine in question for each running process under control. It should also allow any (Unix) signal to be sent to those processes -- I found it necessary to be able to send STOP,CONT, USR1 and USR2 to some of my processes -- INTR and KILL are other natural candidates for "rogue" processes. Yes, I could (and did) rsh a ps and rsh a kill, but this involved a complicated combination of using the monitor to identify a troubled process, figuring out the machine it was on, figuring out its (Unix) PID and (Unix) status, and sending it a (Unix) signal, all the while popping in and out of the monitor or using multiple windows. Although not strictly necessary, it would be wonderful to have a nifty X-client for the monitor program with buttons and data entry lines and not just a command line version. 2) One of the HARDEST problems to solve in contemporary network programming is how to write an numerical application with a non-trivial execution loop (that might run for minutes to hours in a single loop pass or days or weeks in total) and still allow real-time program interruption for modification of its run state. This problem is especially irritating when one is trying to write a program with an X-based front end/graphical display. Current solutions include: a) Just use blocking or non-blocking I/O somewhere inside the main event loop (X or otherwise). This is fine if the execution loop is grained to complete in <.1 sec, terrible if it completes in >10 sec (so one has to wait, and wait, and wait, for the program to get to the I/O statement). Note that I disallow entirely the option of polling in the innermost loops for both efficiency and blocking reasons. b) Separate the front end and the main loop into different programs, called or forked or rsh'd as appropriate. Set up a the I/O inside the short event loop of the control program. Set up sockets and/or pipes and a signal mechanism to asynchronously interrupt the execution module, set a flag to interrupt the innermost loops at a suitable point (a simple conditional is much cheaper than polling for I/O), handle some I/O in a controlled manner, and resume execution, or restart the program, or terminate it after writing out its current state, or whatever. Note that I have figured out how to do all or most of this with standard calls and tools -- but doing it is extremely non-trivial network programming, especially when master (front end) and slave are on different CPU's. You could Make The World A Better Place if you could construct a simple library of function calls that would reduce all the hard part of this (fork/exec/rsh'ing, building the socket I/O channels, establishing the signal mechanism that tells the slave to listen to the master via the channel, and finally making the messaging channel reliable) to boilerplate. Quite a bit of this is already in pvm (one reason I like it) but there's quite a bit missing, especially on the asynchronous interrupt and signal support side (see item 1, for example). As a test case and demonstration, you should build a simple demonstration of an X front end that serves as a mandelbrot DISPLAY (only) with buttons and sliders and all, and a pvm-driven set of clients that run asynchronously and can be interrupted and reset at any time from the front end. X-based mandelbrot programs abound, but the all suffer from the annoying feature that they are not interruptible in the main X event loop until they finish their current picture, unless they poll for I/O in the "wrong place". This can take a LONG time, or waste a LOT OF time, depending on which you choose to do. If you build the support routines needed to do this as a standard part of pvm and make them really easy to use and functional, you could even submit the result to the X-Consortium as a contributed library to be included with the standard release, as it would meet a real need and fix a real shortcoming of X. I know many physicists who have struggled with this particular problem and (usually) worked out their own solution at the cost of reinventing dozens of network programming wheels or (occassionally) just thrown up their hands in disgust and run/interrupted/reset things more simply, more expensively, and less conveniently by hand. 3) Finally, yes, an interactive parallel (x-based) debugger would surely help debug the really hard problems that can arise in parallel code. When last I used your program, I don't think that it had one, although a friend did mention that you have started to provide some sort of facility in this regard. Keep it up. Eventually one would like to be able to single step and debug code running on multiple processors from a single point of development inside some sort of sensibly designed GUI. > ps. Previous survey results are available through World Wide Web at > http://www.epm.ornl.gov/pvm/ Thanks, I'll check it out. rgb Robert G. Brown http://www.phy.duke.edu/~rgb Duke University Dept. of Physics Box 90305 Durham, N.C. 27708-0305 Phone: 1-919-660-2567 Fax: 919-660-2525 email:rgb@phy.duke.edu ------------------------------------------------------------------------------- From: viredaz@research.nj.nec.com (Marc Viredaz) Message-Id: <9501101609.AA00950@iris48> To: pvm-survey@cs.utk.edu Subject: Re: PVM survey >Are you using PVM today? We are not using PVM today, but we are considering it as a target paradigm for a novel MPP, under design at NEC Research Institute. Another model considered is MPI >What are your application(s)? Targeted applications are compute-intensive scientific code. ----------------------------------------------------------------------------- Marc A. Viredaz Phone: (+1 609) 951 2790 NEC Research Institute ? ------------------------------------------------------------------------------- From: rea@unh.edu >Are you using PVM today? Not really. >What are your application(s)? I was using it in a Parallel Computer CS719 class. >How many and what kind of machines have you used with PVM? I used & set it up on Linux (486DX50 20MB running latest kernel) and SGI (R4000 & R4400 32-64MB running IRIX 5.2) >Any comments and suggestions for improvement are welcome. >Please email the information to pvm-survey@cs.utk.edu. A good way to stop pvmd's running on other machines would be good. If I run on 8 sgi's and one dies, it would be nice if the starting up of the pvm console could actaully kill off all the other nodes for me. Instead it seem to not allow you to add the node and you must go kill it. I can see where it might not the the default case but it would be nice if it was an optional startup parameter. >Your help is much appreciated. > >Best wishes, >The PVM Project Team > >ps. Previous survey results are available through World Wide Web at > http://www.epm.ornl.gov/pvm/ > >pps. We apologize if you received multiple copies of this message. >Our first attempt at sending the message uncovered a bug in our mailer. > ------------------------------------------------------------- Robert E. Anderson email: rea@unh.edu Systems Programmer phone: (603) 862-3489 UNH Research Computing Center fax: (603) 862-4623 ------------------------------------------------------------- ------------------------------------------------------------------------------- From: We are currently using pvm in two different Monte Carlo problems. One is a radiation heat problem which has random emittance of "photons". The second is running random realizations of groundwater flow problems. > How many and what kind of machines have you used with PVM? Short answer is 8 SGIs (Irix5.2 & 4.05) and 3 SUNs (SunOS 4.1.1.2) all of which are single processor machines. The long answer: Although we are a visualization lab (data modeling, geometric modeling, video production, etc), two of the staffers are engineers whose primary research focus requires extensive numerical work. We both thought that pvm might be an additional tool for our work. We have pvm running on 3 SUNs and 8 Silicon Graphics machines. All of our machines are connected by ether net although we are considering ATM or FDDI. We are not doing any shared memory pvm programming and do not envision future work in that area. This is because our primary goals are to solve engineering problems, not to make advances in high performance computing. Thus simple calls to start discrete sub-problems on other machines, as the machines become available, is exactly what we want out of pvm. People like us rely upon examples in the pvm documentation to quickly get things going. We do not want to take the time to learn the verbiage of computer experts, and thus we appreciated the examples in the documentation and encourage more examples illustrating the capabilities and tricks of pvm. Randy Fedors rfedors@vis.colostate.edu Civil Engineering Dept < Ph.D. student Groundwater Program > (303)491-8202 Computer Visualization Lab < Visualization Consultant > (303)491-1578 Colorado State University, Fort Collins, Colorado 80523 ------------------------------------------------------------------------------- From: I am using PVM today, on a cluster of workstations (DEC stations with ULTRIX, HP stations with HP-UX, SUNs with SUNOS) and a pentium with Linux. In total there are more than 80 stations, but most of the time I only use 10 or 20 of them. My applications are cryptanalytic attacks. Some of the most successfull attacks on for instance the DES use lots of identical operations, easy to distribute over lots of processors. The problem I have is that sometimes I lose a machine from the cluster, and that not all the hosts are aware of this: if I start with 50 hosts, after a night I typically lose 1 or 2 hosts (that is due to network problems here), but some of the other 48 hosts think they are still with 50. The 2 lost hosts sometimes think they are still in the group. Perhaps that could be improved ? Vincent ------------------------------------------------------------------------------- From: "Robert G. Brown" > Any comments and suggestions for improvement are welcome. Addendum to previous note: a) Distribute the software directly through the Web page you advertised. Why force potential users to go through netlib? b) Your web server seems sluggish. Move the primary pvm distribution off onto an SS20 or something, mirror it on the (possibly overloaded) netlib server, and make it available via anon-ftp/gopher/WWW as WELL as netlib. Frankly, netlib is an idea whose time has completely and irrevocably passed with the overwhelming acceptance and universality of WWW/Mosaic. The whole netlib distribution should now be made available ONLY through the WWW. Feel free to pass this idea on to its keepers (they probably know it already). c) Looking through the release notes through 3.3.6, it looks like several of the features I suggest are still absent. I suggest that you really look carefully into the signal/non-blocking I/O problem and work out a better solution than currently seems implemented, and better control and debugging features are always welcome... rgb Robert G. Brown http://www.phy.duke.edu/~rgb Duke University Dept. of Physics Box 90305 Durham, N.C. 27708-0305 Phone: 1-919-660-2567 Fax: 919-660-2525 email:rgb@phy.duke.edu ------------------------------------------------------------------------------- From: Atanas Radenski > > Are you using PVM today? No, but I plan to use it in the near future. > > What are your application(s)? An experimental compiler for a parallel OOP language. > > How many and what kind of machines have you used with PVM? DEC Alpha workstations (we have a network of 10). Atanas Radenski Department of Computer Science Winston-Salem State University P.O.Box 13069 Winston-Salem, NC 27110 Phones: (910) 750-2478 (office) (910) 765-5713 (home) Fax: (910) 750-2499 (office) E-mail: radenski@uncecs.edu ------------------------------------------------------------------------------- From: Anjuli S Bamzai Answers to the three questions: Yes sometimes Mainly data parallel computationally intensive problems Silicon Graphics and NeXT machines (upto 15 total but after about 8 the response gets quite slow) ------------------------------------------------------------------------------- From: ynnerman@vuse.vanderbilt.edu (Bo Anders Ynnerman) > > Are you using PVM today? Yes > > What are your application(s)? > Atomic structure calcluations. > How many and what kind of machines have you used with PVM? > Locally we are using 4 IBM PowerStation 370. We also have access to a 16 node SP2. Anders Ynnerman ------------------------------------------------------------------------------- From: Paolo TUNINETTO Hi, I'm a student of Computer Engineering at Politecnico di Torino, For my thesys I'm developing an application for doing symbolic traverse of FSM. My virtual machine is a network of 8 alpha Paolo Tuninetto PS: Excuse me for my poor English ------------------------------------------------------------------------------- From: patton@uabphy.phy.uab.edu (David Clay Patton) Are you using PVM today? yes What are your application(s)? quantum mechanical determination of the electronic structure of materials How many and what kind of machines have you used with PVM? 1 SUN4 and 2 SUN4SOL2 Any comments and suggestions for improvement are welcome. Please email the information to pvm-survey@cs.utk.edu. I still can not get PVM to run correctly on SUNMP machines. It works fine by itself, but when I run it on a SUNMP and then another machine the communication breaks down and I get all kinds of netinput bogus packet from ... messages. I suspect there are still bugs in the SUNMP port. Your help is much appreciated. ------------------------------------------------------------------------------- From: "Alex Matro" Q1: Are you using PVM today? A1: Yes. Q2: What are your application(s)? A2: We are using Monte Carlo simulations to calculate thermodynamic properties of systems where we need to know the Boltzmann distribution at higher temperatures to ensure proper sampling of the configuration space. [See "Reducing quasi-ergodic behaviour in Monte-Carlo simulations by J-walking: Applications to atomic clusters" by D.D. Franz, D.L. Freeman and J.D. Doll, J. Chem. Phys. 93, 2769 (1990)] Instead of calculating such a distribution ahead of time and storing it we run Monte Carlo simulations at the higher temperatures simultaneously with the lower-temperature calculations and use PVM to pass the higher-temperature configurations to the lower temperatures. Q:How many and what kind of machines have you used with PVM? A:We are using four Silicon Graphics Indigo workstations. -- ******************************* Alex Matro * matro@haydn.chm.uri.edu * (401)792-2597 * Department of Chemistry * University of Rhode Island * Kingston, RI 02881-0809 * ******************************* ------------------------------------------------------------------------------- From: Received: from mac-sb.imt-mrs.fr by imt950.imt-mrs.fr (AIX 3.2/UCB 5.64/4.03) id AA13135; Tue, 10 Jan 1995 18:33:56 -0600 Date: Tue, 10 Jan 1995 18:33:56 -0600 Message-Id: <9501110033.AA13135@imt950.imt-mrs.fr> To: pvm-survey@cs.utk.edu Subject: Re: PVM survey >Are you using PVM today? >YES >What are your application(s)? >Teaching parallel computing >Supporting PVM for scientific users of our machines >How many and what kind of machines have you used with PVM? >RS6K,HPPA,CRAY >Any comments and suggestions for improvement are welcome. Improve shared memory implementation Develop a light-weight, non-intrusive xpvm PVM is a easy-to-think, hard-to-write tool, and have a heavy syntax. Why, as it becomes widely used, change for another DIFFERENT and INCOMPATIBLE standard (MPI) ? ___________________________________________________________ Thierry Arpin-Pont INSTITUT MEDITERRANEEN DE TECHNOLOGIE Centre de Calcul Technopole de Chateau-Gombert 13451 Marseille cedex 20 FRANCE tel : 33-91054568 fax : 33-91054343 email : tap@imt950.imt-mrs.fr ___________________________________________________________ ------------------------------------------------------------------------------- From: cheatham@tettares.ucsf.EDU (Thomas Cheatham) > Are you using PVM today? Yes, although not in any production runs (only for development currently) > What are your application(s)? AMBER, a molecular dynamics package > How many and what kind of machines have you used with PVM? SGI's: 4 processor Challenge (attempted, but not working) cluster of indigo2's (2 machines) heterogeneous (indigo2's, TFP, older machines, not working) (standard LAN, no FDDI) HP735's: cluster of 8, standard LAN (no FDDI) > Any comments and suggestions for improvement are welcome. Fix up the group server so that it is more standard in terms of spawning it. I had great trouble with PVM 3.3.3? getting my group operations running until I modified all my paths and placed the pvmgs in accessible places. Modify the PVM console to have the ability to clean up old /tmp files. ------------------------------------------------------------------------------- From: > Are you using PVM today? Yes > What are your application(s)? A multi-agent bank check recognition system > How many and what kind of machines have you used with PVM? Only a network of 3 SUN-sparc10 > Any comments and suggestions for improvement are welcome. 1 - A graphic interface to allow us to design the processes (agents) and the links between them which would generate a C skeleton file to run the processes could be very useful and time saving (especially for somebody like me who is not a specialist of parallel systems but just an user). 2 - Different examples of applications and configurations could also be very usefull. Lionel Montoliu +------- Montoliu Lionel ---------+--------------------------------------+ | Laboratoire d'Informatique (LIX) | Phone: (33-1) 69 33 34 69 | | Ecole Polytechnique | Fax: (33-1) 69 33 30 14 | | F91128 Palaiseau Cedex, FRANCE | e-mail: montoliu@lix.polytechnique.fr| | http://lix.polytechnique.fr/ | | +-----------------------------------+--------------------------------------+ ------------------------------------------------------------------------------- From: "Gregory A. Baker" > > Are you using PVM today? > Yes, we are in the process of writing a library of subroutines applied to the orbit determination problem. They will use PVM to communicate. > What are your application(s)? > Our basic problem is orbit determination. Specifically, we are looking at three different areas. 1) Estimation of a large gravity field ( about 50,000 parameters ) 2) Determination of orbits using GPS data ( tons of observations ) 3) Baseline determination on the earth's surface using GPS data > How many and what kind of machines have you used with PVM? > Cray T3D, HP 735, Sun Sparc 10, SGI Iris, NeXT -- ******************************************************************* * * * Gregory A. Baker * * * * Center for Space Research Phone : (512) 471-5509 * * University of Texas at Austin Email : baker@csr.utexas.edu * * 2901 North IH-35 * * Austin, TX 78722 * * * ******************************************************************* ------------------------------------------------------------------------------- From: neri@di.unito.it (Filippo Neri) Dear Sirs, We are a group of scientists working in Machine Learning. We are currently using pvm3, in one of our application realized as a distributed system. Our application is relative to the concept learning task by using a genetic algorithm approach. We achieve our goal by letting different GA evolve and cooperate by means of a communication network. We are currently using pvm3 on these architectures: SUN4 it works very well! CM5 Here, pvm3 is not as robust as on the SUN4 architecture. We had problem in the memory allocation for the message passing activity. PVM3 was not able to completly de-allocate memory after the receipt of a message. Now we have partially solved the problem, but it still exists for communication between the host and the processors network. Best wishes Filippo Neri Ph. D. Student Dipartimento di Informatica Universita' di Torino ITALY ------------------------------------------------------------------------------- From: simpson@abacus.mwsu.edu (Richard Simpson) > >Are you using PVM today? I used PVM for the first time this summer in a graduate course in distributed systems. It worked very well and I plan on using it the next time I teach the course. This course only occurs every two years. > >What are your application(s)? I justed picked several simple applications such as distributed integration of a polynomial curve. There is a certain amount of learning curve for PVM that the students had to master and in a 5 1/2 week course little time was left for interesting apps. > >How many and what kind of machines have you used with PVM? We used several PCs running LINUX and a micro VAX running ULTRIX all connected together via a net. > >Any comments and suggestions for improvement are welcome. >Please email the information to pvm-survey@cs.utk.edu. I was very pleased to discover this system. It is seldom that a small school like ourselves can afford any type of parallel or distributed software or hardware. > @@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@ @ Beauty surrounds us yet we are Blind!!! @ @ @ @ Richard P. Simpson Work Phone: (817)689-4191 @ @ Assistant Professor @ @ Dept. of Computer Science simpson@abacus.mwsu.edu @ @ Midwestern State University @ @ 3410 Taft @ @ Wichita Falls, Tx 76308 @ @@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@@ ------------------------------------------------------------------------------- From: Sowmini Varadhan > > Are you using PVM today? Yes. > What are your application(s)? Parallel numeric computing for the computation of the SVD of large, sparse matrices. > How many and what kind of machines have you used with PVM? Sparcs, RS6K > Any comments and suggestions for improvement are welcome. > Please email the information to pvm-survey@cs.utk.edu. Some comments on the PVM book, available on-line- The PVM book leaves much to be desired: It's not clear what _kind_ of audience this book is targetted for. Is it intended for 1. the scientist/engineer who wants to use PVM to implement some algorithm in parallel, or 2. the network programmer, who wishes to understand the implementation details? In case 1, description of the implementation details should be intended only to show how efficiency gains may be achieved. In case 2, implementation details should show how the applications motivated PVM design. In either case, Chapter 7 ("How PVM Works") is confusing, and appears to be badly organized; while it makes an attempt to describe the data structures used to implement buffers, host tables etc., there is a recurring problem of cross-references: terms are used before they are clearly defined, so that it neither throws light on the implementation details, nor does it indicate how to use the libraries to obtain max. efficiency. This chapter needs some serious restructuring. --Sowmini ------------------------------------------------------------------------------- From: roy@nersc.gov (Roy Troutman) > Are you using PVM today? > Yes. I use version 3.2 to talk to a Cray T3D and 3.3 on a Sun cluster. > What are your application(s)? > Volume rendering and Ray Tracing. > How many and what kind of machines have you used with > PVM? > 16 processors of a Cray T3D and 28 Sun 2 class machines. ------------------------------------------------------------------------------- From: jesus@sobrino.eui.upm.es (Alonso Segoviano) > Are you using PVM today? Not in this moment, but in the past and soon in the future. > What are your application(s)? It have been of two types: 1- Thermodynamic & Aerodynamics. 2- 'Distributed processing' teaching > How many and what kind of machines have you used with PVM? a- 8 machines b- ibm/6000 (530+580) Regards, Jesus ---------------------------- Jesus Alonso Segoviano Tf. (34.1) 336.78.81 Fax (34.1) 336.75.20 Escuela Universitaria de Informatica - OEI Universidad Politecnica de Madrid. Crta. de Valencia km-7. 28031 Madrid. Spain jesus@eui.upm.es ____________________________ ------------------------------------------------------------------------------- From: mcastro@dsee.fee.unicamp.br (Marcelo Stehling de Castro) >Are you using PVM today? Yes, I am. >What are your application(s)? It's an On-line operation and control of power systems using distributed computing environments. >How many and what kind of machines have you used with PVM? I am using 3 different kind of machines using 2 architetures : SUN4 and RS6K which are described bellow: DCE#1: Sparc2+Ethernet This distributed environment consists of ten SUN UNIX-based workstations (one server 470 and nine Sparc2 model) connected as an Ethernet LAN (coaxial cable), featuring nominal bandwidth of 10 Mbps. DCE#2: RS6K/560+Ethernet It's an homogeneous cluster of eight IBM RISC/6000 workstations (model 560) interconnected as an Ethernet token-ring LAN (coaxial cable) with nominal bandwidth of 16 Mbps. DCE#3: RS6K/370+FDDI The third DCE used for testing is a cluster of eight IBM RISC/6000 workstations (model 370) that constitute a SP1 computer (Scalable PowerParallel System 1). Each node has 256 Mbytes of RAM and delivers peak performance of 125 Mflops. The nodes are connected as a fiber optic token-ring LAN (Fiber Distributed Data Interface/FDDI), running at nominal bandwidth of 100 Mbps, through a