Fri Aug 20 15:35:23 1993 =============================================================================== Subject: pvm survey To: pvm-survey-list = those who have requested PVM 3 src from netlib@ornl. Dear Colleague, In order to improve PVM we are trying to understand how people are using the system. 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 send the information to pvm-survey@cs.utk.edu. Your help is much appreciated. Best wishes, The PVM Project Team =============================================================================== From: Gary Hanyzewski >| >| Are you using PVM today? Personally.. No. Institutionally.. Yes. >| >| What are your application(s)? Distributed Global Change Modeling >| >| How many and what kind of machines have you used with PVM? Networks of Sun Sparcstations, SGI's ( everything from indigos and personal Iris's to Challenge Series multi-processor reality engine machines ), NEXT, DECstation ( 5000 and 3000 ), Cray Y-MP, Intel iPSC/860 and Intel Paragon. >| >| Any comments and suggestions for improvement are welcome. >| Please send the information to pvm-survey@cs.utk.edu. >| Your help is much appreciated. Additional support for an better integration of MPP PVM versions with non MPP code. I am still having some small trouble in getting PVM 3.1 to work correctly on our PARAGON. >| >| Best wishes, >| The PVM Project Team And to you. PVM is a great tool. Gary H. ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ Gary Hanyzewski garyh@sdsc.edu San Diego Supercomputer Center (619) 534-5129 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ ------------------------------------------------------------------------------- From: heistand@iastate.edu Dear Sirs, I am quite happy to fill out your survey as I wish PVM to become as powerfull and widespread as possible. I am using PVM to obtain a numerical solution to a 3-d fluids problem, basically the flow around two aircraft wings in close proximity. I am very impressed and pleased with PVM. I am running the model on DEC's, approx. 50 machines, most of them being 5000/25 and a few 5000/133 and 5000/240's. Steve Heistand ------------------------------------------------------------------------------- From: Roberto.Bisiani@N3.SP.CS.CMU.EDU Are you using PVM today? YES What are your application(s)? Speech recognition How many and what kind of machines have you used with PVM? DEC 3000, up to 5, Ethernet (the application really needs a much faster network that I will get in the future). Any comments and suggestions for improvement are welcome. My application exchanges lots of data. I am often getting into trouble with lack of memory. I am currently frustrated enough that I have started looking at pvm sources to understand what is going on. I only use the direct communication option since communicating through the demon does not make much sense in my case and only worsens performace and memory allocation problems. Roberto Bisiani ------------------------------------------------------------------------------- From: nrl@SSESCO.com > > Are you using PVM today? Yes, as we speak. > What are your application(s)? Meteorological, Photochemical,and Hrdologicall models which heretofor could only be run on LARGE Cray machines > > How many and what kind of machines have you used with PVM? At this site we regularly run on from 2 to 7 RS6000s. At times we use 2 SGI and two SPARC systems. At FSU we have used up to sixteen machines for runs of up to six hours. > Any comments and suggestions for improvement are welcome. > Please send the information to pvm-survey@cs.utk.edu. Still waiting to hear from Al about my temporary implementation of the "speed" parameter in the host data structure... ------------------------------------------------------------------------------- From: wcs@nas.nasa.gov (William C. Saphir) >Could you take a moment to answer three questions. (I am one of the main contacts at NAS for people using (or thinking of using) PVM. The information below includes myself and the people with whom I have regular contact. Does not include people who are just starting (several people are thinking about using PVM for CFD applications) I may have missed a few. Hopefully they are on your list). ----------------------- > >Are you using PVM today? Yes. > >What are your application(s)? >How many and what kind of machines have you used with PVM? 1. NAS parallel benchmarks (tightly coupled. lots of communication) PVM 2.4. Has been run on SGI workstations. Poor speedup. Being ported to 3.1, mpps (?) Contact: Raad Fatoohi (fatoohi@nas.nasa.gov) 2. Medusa (CFD application. uses domain decomposition with overset grids so that communication requirements aren't too large. manager-worker model. Run on SGI workstations. Good speedup. Contact: Merritt Smith. (mhsmith@nas.nasa.gov) 3. Another cfd application. Also manager-worker. Run on Intel Hypercube (and network of workstations?). Good speedup. Contact: Graham Rhodes (sues@nas.nasa.gov) 4. Simple benchmarks to measure PVM/mpp performance vs. native mpp message passing. Contact: me On number of workstations: Order of 10. iPSC/860 nodes: up to 128. > >Any comments and suggestions for improvement are welcome. >Your help is much appreciated. In no particular order and of varied importance. 1. For many sites, instructions (and a simple procedure) on how to install PVM in a central place (not a home directory) would probably be useful. 2. As I've already commented to a few people, support for global group operations would be useful for us. Dynamic groups do not seem to be too useful for numerical codes. This is something I'm working on. I'll keep you informed of progress. 3. I/O could be a real problem for a lot of cfd applications. I realize it is probably impossible for PVM to do anything about this. People who need to do a lot of I/O will either avoid PVM or write machine-specific code to handle it. 4. Is the "f" in fortran calls necessary? It's ugly. PVM_send() and pvm_send() perhaps? (like MPI). 5. PvmDataInPlace would be really good (although I suppose since multiple packs probably result in multiple sends it should come with a warning). Also, xdr encoding only when necessary would be good. Several people have mentioned these. 6. The FAQ that's floating around is out of date and not very good. 7. Any plans for an instrumented library (or a flag?) to produce traces? 8. A utility library would be useful. Timers are the main thing. I'm planning to produce one, and will keep you informed. >Please send the information to pvm-survey@cs.utk.edu. 9. The survey was sent from "owner-pvm-survey." Although replies go the right place, you may lose some: telnet cs.utk.edu 25 Connected to cs.utk.edu. 220 CS.UTK.EDU Sendmail 5.61+IDA+UTK-930125/2.8s-UTK ready at Tue, 17 Aug 93 22:25:41 -0400 vrfy owner-pvm-survey 250 ---------------------------- You're probably aware of this, but we seem to have 2 distinct categories of PVM users. 1. People making use of idle workstation cycles. (Cheap powerful computing). 2. People who want their codes to be portable. (mostly MPP users). There's a gray area, of course, but it often seems like one reason is dominant. Bill Saphir wcs@nas.nasa.gov ps. I'd be interested in seeing the results of this survey. ------------------------------------------------------------------------------- From: kortas@ctrss1.lanl.gov (Samuel Kortas) yes, we are using PVM. We have a global ocean code here in Los Alamos, developped on the CM-5, and our job is to make it run on other MIMD configurations. Among those, we made it run on an IBM Cluster (1 to 16 nodes), under PVM 2.4, and on a T3D emulator under PVM 3. All our codes are written in fortran (either 77 or CM-Fortran) > Any comments and suggestions for improvement are welcome. > Please send the information to pvm-survey@cs.utk.edu. > Your help is much appreciated. Yes, PVM is great because we hope that in a long run we'll only need one version of our code for the IBM cluster, but also for the CM-5, the IPSC or the Paragon. But... we had a small problem with the length of a message containing integers. the instructions fputnint and getnint don't seem to work the same everywhere : we always work in double precision (INTEGER*4) and on certain machines, (the IBM Cluster for example), if we want to send/receive and integer message of size n we need to call fputnint (buffer,n*2, info). on other just the regular fputnint(buffer,n,info) will work.... Have you heard of other persons who might have had the same problem? fputnint really sends INTEGER*4 doesn't it? > Best wishes, > The PVM Project Team Thank You for providing a portable environement! |\/\/\/| --------------------------------------------------- | | Samuel Kortas kortas@lanl.gov | | M.S K723 | (o)(o) LOS ALAMOS NATIONAL LABORATORY C _) LOS ALAMOS NM 87544 | ,___| UNITED STATES | / /____\ GRA / T-3 Phone : (505) 665-7782 / \ --------------------------------------------------- ------------------------------------------------------------------------------- From: chuckr@bigbird.llnl.gov (Chuck Rendleman) > > What are your application(s)? Computational fluid dynamics, adaptive mesh refinement. > > How many and what kind of machines have you used with PVM? Massive Parallel systems (CM5, Meiko CS2, Cray T3D, Paragon). Development work on SGI crimson. ------------------------------------------------------------------------------- From: Bill Pearson Are you using PVM today? Extensively What are your application(s)? DNA and protein sequence comparison How many and what kind of machines have you used with PVM? Sun4s: 16, SGI-Indigos: 6, DecALPHA's: 5 ------------------------------------------------------------------------------- From: eric@acis11.tamu.edu Are you using PVM today? Yes we are. What are your application(s)? We are using it to compare network traffic/performance over ethernet, fddi, and token-ring. How many and what kind of machines have you used with PVM? currently, we have 5 IBM rs/6k's (4 320h & 1 550) and 4 sparc-10's (1 4/50 and 3 4/75). Comments: We've just started this work and as yet do not have the fddi/tr up yet. I've heard rumors about trouble with mutilple interfaces and only hope they aren't true???? Hope this helps. Eric ---------- || Eric L. Nelson (Eric.L.Nelson@tamu.edu) _/_/_/_/ || Research Associate _/ _/_/_/ _/ _/_/_/ || Autonomous Underwater _/_/_/ _/ _/ _/ _/ _/ || Vehicle Controller Project _/ _/_/_/ _/ _/ || Texas A&M University _/_/_/_/ _/ _/ _/ _/_/_/ || (409) 845-0085 / (713) 256-1528 ------------------------------------------------------------------------------- From: brown@salt.ra.anl.gov Yes, I am using PVM. Several other poeple in the Reactor Analysis Division are also using it for different applications. (I have no info on PVM usage for the rest of Argonne.) At present, we are still using version 2.4, but expect to switch to 3.1 sometime soon. My applications: Monte Carlo calculations for neutron transport & reactor analysis. Other applications in the Division: Statistical analysis of experimental data; analysis of measured data related to fuel-failure indicators. Planned applications: Neutron diffusion theory. +--------------------------------------------------------+ | Forrest B. Brown || email: fbbrown@anl.gov | | Argonne National Laboratory || voice: (708) 252-9622 | | 9700 S. Cass Avenue - RA/208 || fax: (708) 252-4500 | | Argonne, IL. 60439-4842 || | +--------------------------------------------------------+ ------------------------------------------------------------------------------- From: prange@sdr.slb.com (Michael Prange) > Are you using PVM today? Yes. > What are your application(s)? Numerical modeling of elastic and EM wave propagation. Modeling neutron transport. Solving large matrix systems using blacs. > How many and what kind of machines have you used with PVM? 40 workstations over ethernet and internet. Sparc1+, Sparc2, Sparc10, RS6000, Iris Indigo. I'd like to use our dec alphas running VMS. > Any comments and suggestions for improvement are welcome. > Please send the information to pvm-survey@cs.utk.edu. > Your help is much appreciated. > Keep up the good work. I'd like to see tools to simplify load balancing. Also, hence is a nice idea, but it needs to include load balancing and more generality. Michael ------------------------------------------------------------------------------- From: rdaniel@acl.lanl.gov (Ronald E. Daniel) > Are you using PVM today? Yes, but I have just started working with it in the last month. > What are your application(s)? My primary interest is in debugging parallel programs. The parallel debugger itself does not use PVM, although I am using Xab3 beta to collect events for tracing execution. For mostly-unrelated work I will be using PVM to implement a constraint satisfaction algorithm based on simulated annealing. > How many and what kind of machines have you used with PVM? Currently, only Sun workstations. In the not-too-distant future I will probably go to a heterogeneous mix of machines - IBM, SGI, and Sun workstations. Ron Daniel Jr. email: rdaniel@acl.lanl.gov Advanced Computing Lab voice: (505) 665-7453 MS B-287 TA-3 Bldg. 2011 fax: (505) 665-4939 Los Alamos National Lab tautology: "Conformity is very popular" Los Alamos, NM, 87545 ------------------------------------------------------------------------------- From: Mark Bolstad Message-Id: <9308161805.AA25464@blad.rtpnc.epa.gov> To: pvm-survey@cs.utk.edu Subject: Re: pvm survey > Are you using PVM today? Yes. > What are your application(s)? Volume Rendering, Data Interpolation, Ray Tracing, Radiosity > How many and what kind of machines have you used with PVM? 11 Machines with 22 Processors (1 each of an 8,4,2 processor machines) SGI Dec Sun Cray Data General I would like to see a port to the MasPar. Mark Bolstad ------------------------------------------------------------------------------- From: htree@gdstech.grumman.com (Hong-Yi Ip) Are you using PVM today? Not today, but may resume in the future. What are your application(s)? image processing, distributed database, etc... How many and what kind of machines have you used with PVM? Suns, DEC ultrix, SGI... Any comments and suggestions for improvement are welcome. Please send the information to pvm-survey@cs.utk.edu. Your help is much appreciated. Best wishes, The PVM Project Team Please send me a summary of this survey if possible. regards --Hong-Yi ***************************************************************************** Hong-Yi Ip (516)346-8724, FAX:(516)575-0965, email: htree@gdstech.grumman.com Grumman Data Systems, addr: MS D12-25, 1111 Stewart Ave, Bethpage, N.Y. 11714 ***************************************************************************** ------------------------------------------------------------------------------- From: guo@meindigo.esrd.com (Guo) >Are you using PVM today? Yes. >What are your application(s)? p-version finite element methods. >How many and what kind of machines have you used with PVM? 4 machines: SGI, RS6K, SUN4, HPPA >Any comments and suggestions for improvement are welcome. NCUBE porting would be useful to us. ------------------------------------------------------------------------------- From: jeg@kaiser.Lanl.GOV (James E Gubernatis) > > Are you using PVM today? Yes. > What are your application(s)? Qunatum Monte Carlo simulations exact diagonalization studies of systems of strongly interacting electrons. > How many and what kind of machines have you used with PVM? Separately on a network 6 Sun workstations and a network of 16 IBM workstations. ------------------------------------------------------------------------------- From: chih@MasPar.COM (Sam Chih) > Are you using PVM today? > Since we received the PVM, we haven't completed the port to our MPP machine yet. > What are your application(s)? > Not decided at this time. > How many and what kind of machines have you used with PVM? > We have not used the PVM yet. > Any comments and suggestions for improvement are welcome. > Please send the information to pvm-survey@cs.utk.edu. > Your help is much appreciated. > Can you periodically publish the current development or user experience with the PVM ? Thanks. ------------------------------------------------------------------------------- From: Jamie Painter Are you using PVM today? Yes > What are your application(s)? Volume Rendering of 3D Scalar data sets. The volume rendering code originated on a CM-5 using CMMD for message passing and was ported to PVM to allow use in a workstation network environment. > How many and what kind of machines have you used with PVM? We've run up to 32 nodes with the following architectures: HPPA, RS6K, SGI, SUN4. --Jamie ------------------------------------------------------------------------------- From: "Brian Grant" Reply to: RE>pvm survey Yes, I am using PVM3.0 for "coarse-grain decomposition" in solid-state physics codes. For example, if I need to solve for wave functions at numerous k-points, I will use pvm to spread them out. Because of network latency, I do not use fine-grain decomposition. The machines I use, RS6K's, are so unevenly balanced, I must use a queue. This conflicts with the latency (request-response time) and, inorder to see appreciable speedup, I must initialize with 2 jobs per machine. This allows "asynchronous messages". J. Brian Grant ------------------------------------------------------------------------------- From: djc@dsmail.hmi.de (Markus Dahm) Hi, > Are you using PVM today? Probably :) I'm working on an comparison report between EXPRESS and PVM; and as far as I can see PVM wins the race. > What are your application(s)? Mainly we are trying to parallelize an FORTRAN(Yuk) application calculating cluster configurations of gas atoms. > How many and what kind of machines have you used with PVM? 9 SPARCS, 2 VAXes and a 4 processor CONVEX. I am quite fond of the PVM programming environment, it even may be that I will write my graduation using PVM. Yours sincerely --- __ ______ /\ /\ /\ \ | ___ \ Markus Dahm /\ \/\ \ / \ \ | | | \ \ / \/ \ \ / /\ \ \ | | | | ) Voice : +49 (0)30 805 61 34 / /\ /\ \ \ / / /\ \ \ | | | | | Office : +49 (0)30 8009 - 2594 / / /\/_/\ \ \ / / /--\ \ \ | | |__/ ) E-Mail : markusd@cs.tu-berlin.de / / / \ \ \/ / /----\ \ \ | |/__/ / news@dsmail.hmi.de /_/_/ \_\/_/_/ \_\_\|______/ djc@dsun3.hmi.de ------------------------------------------------------------------------------- From: Tom Bartol Hi, I use pvm 3.1 a lot. I'd use it more if the folks over here at the San Diego Supercomputer Center could get it working properly on their Paragon. I've written a Monte Carlo computer simulation of synaptic function in the nervous system. I've use pvm on a hetergeneous network of sparcs, SGI's, and RS/6000's. I'd like to be able to use it on the Paragon machine here but they've been having trouble getting it working. I really like pvm. Thanks for your wonderful efforts on this product. It's a great alternative to EXPRESS. Thanx again, Tom ------------------------------------------------------------------------------- From: Kensuke Shirakawa Q. Are you using PVM today? A. Yes, I am. Q. What are your application(s)? A. I am currently testing PVM 3.1, to see the relationship between the number of hosts in the virtual machine and the speed of a distributed matrix multiplier program. Q. How many and what kind of machines have you used with PVM? A. I have been using up to 8 DEC 5000 workstations connected by an internet. ------------------------------------------------------------------------------- From: louis@arthur.ee.up.ac.za (Louis Coetzee) Are you using PVM today? Yes. What are your application(s)? Parallel image processing algorithms. How many and what kind of machines have you used with PVM? IBM RS6000: 1*340 3*220 A pleasure. Louis Coetzee University of Pretoria, Pretoria, South-Africa ------------------------------------------------------------------------------- From: weimar@eole.ulb.ac.be (Joerg Richard Weimar) Are you using PVM today? yes. Also yesterday. In fact, I use it every day since it has become my standard interface between calculation and display. What are your application(s)? Cellular Automata, lattice gases. How many and what kind of machines have you used with PVM? Workstations: 1-4 Type: NEXT, SGI, SONY, SUN Any comments and suggestions for improvement are welcome. Thanks a lot for this tool. One thing that I am waiting for is the implementation of the InPlace data type (maybe it is already done, I haven't upgraded for some time (still 3.1 no patches)). One problem I have with upgrading is that the modifications I made to directories to support one of our workstations (add machine type SONY), got destroyed when I upgraded from 2.4 to 3.1. Since everything is running somewhat stable right now, I am reluctant to do that work again. -- Joerg R. Weimar #!###!!!##!!#!#!!!###!!##!!##!!!###!!!##!!##!#!#!#####!#!##! jweimar@ulb.ac.be (NeXTMail welcome) Service de Chimie Physique II, Universite Libre de Bruxelles #!##!##!#!#!#!!!##!!#!#!#!#!##!##!#!##!#!##!##!!#!#!#!##!#!# ------------------------------------------------------------------------------- From: paia@chert.CS.ORST.EDU (Akash L PAI) > Are you using PVM today? YES > What are your application(s)? NUMERICAL ANALYSIS, TOOLS FOR PVM > How many and what kind of machines have you used with PVM? USED PVM ON RS6000, TRYING TO INSTALL IT ON THE i860 ------------------------------------------------------------------------------- From: jjv5@retina.chem.psu.edu (Jim Vincent) Are you using PVM today? yes What are your application(s)? parallel implementation of a molecular mechanic/dynamics package written by someone else (in Fortran), that is widely used How many and what kind of machines have you used with PVM? SGI Indigo, 4D IBM RS6000 IBM SP1 Sun 4 series Suggestions: To get feedback for the sake of improvement how about asking what was/is the most difficult part about using PVM? My answer: Installing and using on many machines. It isn't that difficult but it is a pain in the but to copy and install over and over again. I have made self install scripts for installing on reomte machines from a master and also makefiles that copy source code and compile on all machines in a hostlist from a master machine. This greatly simplifies using PVM on a wide scale. ------------------------------------------------------------------------------- From: spector@gandalf.Berkeley.EDU (Phil Spector) >> Are you using PVM today? Yes >> What are your application(s)? Parallelization of the CART (Classification and Regression Trees) algorithm. >> How many and what kind of machines have you used with PVM? A network of 40 Sun Microsystems SparcStation 1+ 's. - Phil Spector Statistical Computing Facility Department of Statistics UC Berkeley spector@stat.berkeley.edu ------------------------------------------------------------------------------- From: kbarnes@hannibal.atl.ge.com (Kerry B Barnes) >Are you using PVM today? Yes >What are your application(s)? We are using it as the communication layer for our Graphical Entry Distributed Application Environment (GEDAE) that allows a user to build applications as data flow graphs and them map those flowgraphs around a heterogeneous network of processors. PVM was used to reimplement the communicaton layer of GEDAE giving us the benefit of having the system dependent communication code being automatically ported to a large number of processor types. How many and what kind of machines have you used with PVM? We have only used PVM on sparkstations, though we have had GEDAE (with the old communication layer) working on Vaxes running Ultrix and VMS, and on a several other processors. Being able to have flow graphs working in a heterogenous environment is important to us. Any comments and suggestions for improvement are welcome. Would very much like a multithreaded version of PVM, perhaps based on POSIX standard. This would allow us to avoid some of the complex workarounds we need to do to get asynchrony using unix signals, but without a "signal safe" unix library. In fact though we currently have this working reliably in our old communication layer (tested quite exhaustively over a period of years) we had to avoid using unix interrupts with PVM. Therefore though the current version of PVM gives us the portability we desire it does not allow us to provide the responsiveness availble in our current layer where processes can send each other preemptive commands. ------------------------------------------------------------------------------- From: Mike Norman I am currently not using PVM. I intendt o use PVM in the near future for a Vehicle Routing Application. Hardware is still undefined. MIke ------------------------------------------------------------------------------- From: Victor Eijkhout 1/ Yes I use pvm regularly 2/ Sparse iterative matrix solvers 3/ At most 8 hosts, Sun4, HPPA, RS6K. Victor Eijkhout Department of Computer Science; University of Tennessee at Knoxville 107 Ayres Hall; 1403 Circle Dr.; Knoxville TN 37996-1301; USA phone: +1 615 974 8298 (secretary 8295; fax 8296); home +1 615 558 3069 Support the League for Programming Freedom! lpf@uunet.uu.net ------------------------------------------------------------------------------- From: Frank.Pinski@UC.Edu (Frank Pinski) Yes. Some test problems. 15 Sun SPARCS Question: Is possible to spawn jobs with a nice value? (The procedure in the manual doesn't seem to work.) Frank Pinski ------------------------------------------------------------------------------- From: petri@ibr.cs.tu-bs.de (Stefan Petri) We have installed it at our institute but are rarely using it. Survey> What are your application(s)? We are investigationg load balancing in distributed systems, and look at pvm because it is a system that the engineering faculties here make heavy use of. Survey> How many and what kind of machines have you used with PVM? 25 Sun Sparc Stefan ------------------------------------------------------------------------------- From: merlo@vega.EPM.ORNL.GOV (Alessandro Merlo) Nowadays I am not using PVM. I used it a little bit last year connecting together an nCUBE with 16 procesors and a network of workstations (SUN3/50, SUN SPARCstations, DIGITAL DECstations). The applications I tested were matrix multiplication algorithms. Since I am developing a tool for workload characterization of parallel systems, and this toll will be able to analyze tracefiles collected using the PICL libraries (among others), I think I will use PVM & HENCE again in the next future. Best regards, Alessandro Merlo Universita' di Pavia Dipartimento di Informatica e Sistemistica Via Abbiategrasso, 209 I-27100 PAVIA ------------------------------------------------------------------------------- From: Roger M. Firestone We are not yet ready to use PVM; it has just been ported over to our operating center and has yet to be compiled. There are aspects of the documentation that are somewhat less transparent than others. I am perusing comp.parallel.pvm to learn about others' experience before plunging in. A possible first application will be for an e&m code for which portability of the resulting software will be a major consideration. We'll probably be using Intel, TMC, Sun, and NeXT equipment with PVM. Sorry I can't be more specific yet. Roger M. Firestone ------------------------------------------------------------------------------- From: Philippe Meunier > Are you using PVM today? No, i am using p4 because i need shared memory message passing on a KSR machine. I was told PVM will soon be able to use KSR's shared memory, so if i have time i will perhaps try to use it. > What are your application(s)? global optimization applied to chemistry. > How many and what kind of machines have you used with PVM? A network of 20 Sun machines. > Any comments and suggestions for improvement are welcome. Some asynchroneous message passing features would be usefull (something like the hrecv() function on the i860 machine). Philippe ----------------------------------------------------------------------- Philippe Meunier meunierp@anchor.cs.colorado.edu ------------------------------------------------------------------------------- From: sienicki@allegra.att.com (Jim Sienicki) Yes, I am using PVM today. I am running an automatic test-pattern generation program on several SUN workstations, using the fault parallelism paradigm. I have run the program on up to eight processors. No problems so far. Jim Sienicki ------------------------------------------------------------------------------- From: wka@cs.duke.edu (Bill Allard) I am not using PVM today. I was using it quite a bit two months ago. When I used it I programmed elementary linear algebra, such as matrix multi- plication, in order to get used to message passing and parallelism; PVM was very helpful in this regard. I used PVM on 25 sun4 workstations in the Duke University Mathematics Department. Bill Allard ------------------------------------------------------------------------------- From: mar@sg6.chem.upenn.edu (Wen Mar) I am using PVM today. My application is Molecular dynamics simulation. I typically use 3-4 machines in my project. They are IBM RS6000, SGI Indigoes, Kendall Square multiprocessor machines. Wen Mar Dept. of Chemistry Univ. of Pennsylvania ------------------------------------------------------------------------------- From: hsu@crl.dec.com Are you using PVM today? Yes What are your application(s)? Volume Rendering How many and what kind of machines have you used with PVM? 4-8 machines. DECstations, soon Alphas. ------------------------------------------------------------------------------- From: jjv5@retina.chem.psu.edu (Jim Vincent) Are you using PVM today? yes What are your application(s)? Parallel implementation of molecular modelling package How many and what kind of machines have you used with PVM? SGI Indigo, 4D IBM RS6000 IBM SP1 Sun 4 series ------------------------------------------------------------------------------- From: wentoh@CS.UCLA.EDU (Wen-Toh Liao) > Are you using PVM today? Yes. Well, sort of. At least not everyday. > What are your application(s)? I am developing and distributing a distributed simulation language that uses PVM as the lower level network communication support. > How many and what kind of machines have you used with PVM? Mainly, SUN4, SPARC, and DEC-station. wentoh ------------------------------------------------------------------------------- From: Soeren Olesen > Are you using PVM today? Occasionally. > What are your application(s)? Simulated annealing, image restoring. > How many and what kind of machines have you used with PVM? 1-32 SUN4 ------------------------------------------------------------------------------- From: V Varadarajan Are you using PVM today? Not today. But I am doing some developement using PVM 3.1. We are exploring its usefulness in distributed computing in electromagnetics. What are your application(s)? Mainly applied numerical analysis. Solution of iterative methods in linear systems and time evolution problems for large problem sizes. How many and what kind of machines have you used with PVM? DEC, Sparc and HP workstations. A few other comments: PVM calls increase the code size tremendously. I think you could provide some routines that can do all the work of pvm_initsend, pvm_pk* and pvm_mcast or pvm_send in one call to it. This will greatly help us keep the code size under control. You should also make provisions for symmetric multiprocessing in shared memory environment; its time will eventually come, perhaps soon. Rajan. u6532@c.nersc.gov ------------------------------------------------------------------------------- From: William T. Rankin Dear PVM Project Team. Firstoff, I'd like to thank you for a job well done with PVM. Here is you survey response. > Are you using PVM today? As we speak. > What are your application(s)? I am using PVM to compute a large numerical sequence called "Golomb Rulers". This forms the basis for my Masters Thesis work, which should be completed in a couple months. The total time required for this computation has been estimated in the ten's of thousands of CPU hours. The processing model I am using a a basic master/slave configuration, using a "processor pool" strategy for load balancing. > How many and what kind of machines have you used with PVM? Currently I am running in a semi-homogeneous network of various models of DecStations (from 2100's through 5000 series machines). At one point I was utilizing 92 processors. I have also run on a small network of 12 Sparc Classics under Solaris 2.2. > Any comments and suggestions for improvement are welcome. I have some comments and suggestions on the Solaris port, which I will post in the near future. thanks again, -bill rankin Duke University wrankin@ee.duke.edu ------------------------------------------------------------------------------- From: Michel Proulx > Are you using PVM today? Yes, we are. We are still in the development phase. > > What are your application(s)? > We are trying to parallelize the EGS (Electron Gamma Shower) code -- Monte-Carlo calculations -- with minimal modification of the existing code (Fortran), with dynamic load balancing, automatic task/host failure recovery, and dynamic addition and removal of participating hosts. > How many and what kind of machines have you used with PVM? > For now, only Sun workstations, but Silicon Graphics machines will be included soon. > Any comments and suggestions for improvement are welcome. > Please send the information to pvm-survey@cs.utk.edu. > Your help is much appreciated. The inclusion of the "pvm group" task in the pvm daemon would be nice. It would give better robustness to the pvm virtual machine. (since the "group" task can be started on a different machine than the master pvm, if either machine dies, the pvm task are either killed or hang...) -- Michel Proulx, National Research Council Canada, Ottawa, Ontario ------------------------------------------------------------------------------- From: derek@crl.dec.com 1. Are you using PVM today? Yes. 2. What are your application(s)? We are using PVM to start up processes on a farm of workstations to act as a multiprocessor. PVM is also used to pass initial data and collect final data. 3. How many and what kind of machines have you used with PVM? 4 machines or few are currently being used with PVM. They are all MIPS-based DEC workstations at the moment. Derek Chiou ------------------------------------------------------------------------------- From: vavasis@cs.cornell.edu (Stephen Vavasis) (1) I used PVM last semester (1/93-5/93). (2) My application was educational -- I was teaching a course on parallel scientific computing. (3) We used PVM on RS6000's -- generally a cluster of eight. -- Steve Vavasis (vavasis@cs.cornell.edu) ------------------------------------------------------------------------------- From: dave@meaddata.com (Dave Pavey) We are currently trying to understand in a client/server/distributed object design, the impact of various couplings between cooperating systems. In effect the difference loose, tight, or as in the case of PVM distributed coupling. We are not currently running PVM, but are rather studying it from the perspective the effect of SMP vs loose coupling and RPC vs Distributed Objects on such a distributed coupling system. This understanding will form a basis for evaluating hardware selections in the future. Our commitment is firmly entrenched with the OSF/DCE,DME technologies and so there is a problem introducing PVM into our architecture. Where technologies are not required to be DCE implementations, they are required to interoperate with such DCE technologies as makes sense. The PVM distribute compute model has significant appeal, but unitl it can be integrated into the DCE environment we won't be able to do anything but investigative inquiry. Dave Pavey Sr. Staff Sys. Eng. Mead Data Central ------------------------------------------------------------------------------- From: rodney@oce.orst.edu (Rodney James) > Are you using PVM today? Yes. > What are your application(s)? 1. Distributed Monte Carlo evaluation of Feynman path integrals. 2. Ocean/atmosphere modeling on workstation cluster 3. Global atmosphere modeling on coupled parallel supercomputer and workstation cluster. > How many and what kind of machines have you used with PVM? RS/6000, HP9000, SUN Sparc, CM-5, CM-200. As many as 32 machines used at once, mostly RS/6000 and SUN4's. > Any comments and suggestions for improvement are welcome. I am very happy with pvm. No complaints, keep up the good work! -- -Rodney James (rodney@oce.orst.edu) Oceanography / Oregon State University ------------------------------------------------------------------------------- From: shaffer@phyast.pitt.edu (C. David Shaffer) Are you using PVM today? Yes. What are your application(s)? I use PVM primarly for number crunching. Currently I have two large applications running under PVM. The first calculates "matrix elements" for inelastic scattering of photons from bound (atomic) electrons (Compton scattering). This calculation has migrated across various parallel platforms (including the BBN Butterfly at LLNL, and a CM5 which used to be at the Pittsburgh Supercomputing Center) but I hope to be able to stick with PVM. The majority of the time consumed by this code is in performing numerical integrals and computing Clebch-Gordon coefficients. It was parallelized at the matrix element level and uses a 'Slave-host' configuration for dynamically assigning a matrix element to be computed by each slave. The number and size of the messages passed are small so this application works well over a busy network. The second application is calculating elastic scattering of photons from bound electrons. This calculation is still being parallelized however we have a version that is parallelized at the level of the atomic subshell from which the photon is being scattered. How many and what kind of machines have you used with PVM? I use 10 workstations including Sun SPARCstation 10, 2, ELC and HP 9000/730. I have access to a few MPP machines (including a CM5 now with NSC) which I hope to integrate into my 'virtual machine'. I have spent the summer at LLNL (in Livermore CA) and am trying to convince many of my peers that using PVM in parallel software development would represent a step forward, away from the machine dependent coding of the past, for our applications. Although the computing needs at the Lab may vary from division to division, here in L-Division, many applications can be parallelized with minimum changes and at a sufficient granularity as to keep message passing at a minimum. I think that such applications are well suited for PVM and hope to convince others. You have provided an excellent package for parallel programming which I hope will save much needed gov't time and money. Keep up the good work! ************************************* The opions expressed here are my own. I do not speak for Lawrence Livermore National Lab or the DOE. ************************************* ------------------------------------------------------------------------------- From: K P Wang > Are you using PVM today? Yes, I am using PVM. > What are your application(s)? Parallel numerical computation: parallel finite difference, parallel finite element, domain decomposition, and load balancing schemes. > How many and what kind of machines have you used with PVM? I am using clusters of SUN4 and SUN3. > > Any comments and suggestions for improvement are welcome. > Please send the information to pvm-survey@cs.utk.edu. > Your help is much appreciated. 1. I wish I could have of parallel I/O functions similar to the I/O in Express such that my read and write can be performed through the same consol. 2. I always have a hard time to understand the error messages and don't known what happen to my program. If there is a list of error messages and their cure then my life will be easier. 3. Thank you too. Kai-Ping Wang ------------------------------------------------------------------------------- From: Chas J Williams III 1. yes 2. molecular dynamics, weather 3. 15 machine rs/6000 cluster 7 machine cluster (3 hp 9000, 1 sgi, 1 sparc 10, 1 alpha, 1 rs6000 ) --chas iii chas@ra.nrl.navy.mil ------------------------------------------------------------------------------- From: Antoine Petitet Dear PVM Project Team, No, I am not using PVM today. My application basically uses PVM to run a program on a remote machine, sending the results back to my workstation. The remote machine is for the moment typically an Intel iPSC/2, iPSC/i860 or a IBM RS6K. It should include the CM-5 in the near future. Remark, it is difficult to install PVM on the I860 for example. Moreover, I didn't see any difference in the documentation and the makefiles to install PVM on the host machine of the hypercube (a PC 386) and the nodes (I860) themselves. In this particular case, I understand that the librairies for the host and the nodes should be stored in distinct directories, which is unclear in the documentation I got from xnetlib. (PVM vers. 3.1). Antoine Petitet ------------------------------------------------------------------------------- From: Hanebutte Ulf Rainer Are you using PVM today: Currently not for everyday simulations, the use of APPL has shown to give a more robust distributed environment. But keep PVM2.4 and 3.1 current. What are your application(s)? Aeroscience Applications, in particular adaptive mesh refinement algorithms for complex shock wave phenomena. How many and what kind of machines have you used with PVM? 16 SUN elc's 8 SUN Sparc 10 or 3 SGI normally as a homogenious cluster ------------------------------------------------------------------------------- From: haque@rcg.cc.iastate.edu 1. Yes, I will try to use it today. 2. Benchmarking. 3. Comments: Wish it had some sensible instructions for installing and startup. ------------------------------------------------------------------------------- From: Vasudha Govindan Are you using PVM today? No What are your application(s)? N-Body simulations Soln of Laplace eqn programs to measure ommunication times, etc. How many and what kind of machines have you used with PVM? 17 SUN workstations (SUN4s, Sparc-1,2,10) (full configuration) I normally use upto 8 workstations to test/debug programs. ------------------------------------------------------------------------------- From: "E. Loren Buhle, Jr. [(215)-662-3084]" >Are you using PVM today? Yes. > >What are your application(s)? 3 dimensional radiation therapy simulation calculations for photons and electrons. Some 3D medical imaging . . . >How many and what kind of machines have you used with PVM? I use Stardent (formerly Ardent), NeXt, DECstation 5000/240, 5000/200, and a DEC AXP/OSF 3000-500. There are two 5000/240s and one 5000/200. >Any comments and suggestions for improvement are welcome. >Please send the information to pvm-survey@cs.utk.edu. >Your help is much appreciated. I'm just getting started, but appreciate your survey and would be curious to see the results.... -- Dr. E. Loren Buhle, Jr. INTERNET: BUHLE@XRT.UPENN.EDU University of Pennsylvania School of Medicine Phone: 215-662-3084 Rm 440A, 3401 Walnut St., Philadelphia, PA 19104-6228 FAX: 215-349-5978 ------------------------------------------------------------------------------- From: Jeff Johnson > Are you using PVM today? > Yes, we are using PVM, but not in full production. Our primary application is still being tested and certified. > What are your application(s)? > Our major application is an in-house CFD code > How many and what kind of machines have you used with PVM? > Hewlett Packard 9000-715,720,730,735 and 755. Currently 7, potentially over 200 within the next 6 months. Silicon Graphics Crimson (1), Indigo (7) IBM RS/6000-320 (1). > Any comments and suggestions for improvement are welcome. Getting a 'pvm' started after a previous failure can be a real pain to an end user. If a previous 'pvm' is still running or if the 'pvmd.uid' file is still laying around, manual intervention is required. This is not acceptable if it is a batch job that is being started in the middle of the night!. A couple of things could help the situation. 1) Create a mode that doesn't require daemons on the remote machines. This would cause slower startup of tasks, but in our case this is not a problem since we start a task and it runs for hours. The 'p4' library from Argonne defaults to 'no-daemon' mode. Note that this eliminates the problems with undeleted 'pvmd.uid' files and old 'pvmds'. 2) Better checks for stray 'pvmd.uid' files. It probaby isn't necessary to bomb just because a 'pvmd.uid' file exists when trying to start 'pvmd'. If a 'pvmd.uid' file exists, try to connect to the pvmd. If the connect is unsuccesful, one can probably assume that the pvmd isn't really running. 3) The master application can't tell when the master 'pvmd' has started. We start 'pvmd' externally to the application with a host file. We had to put a 'sleep' command after the 'pvmd' command to allow it to start before trying to start the application so the application could successfully connect. The time to get started varied widely depending on system load. I mailed a message to the developers, who responded that I could use 'the start_pvmd(?) call from the application. This function is not documented and is not available from FORTRAN (see below). Also, we need full capability from FORTRAN (some very useful functions are only available from C). In particular, 'pvmfconfig' and 'pvmfspawn'. Also, another nasty...ALL externally visible names should have a name that starts with 'pvm' to avoid name space conflicts. Someday, somewhere, somebody will have to link with the pvm library and some other third party library which they have no control over and will get duplicate symbols! Not to sound too critical, 'pvm' has made my life a lot easier. The documentation is good (I've only had to look at the code a couple of times), and responses to mail messages to the developers have been very prompt! Hope this has been constructive! -- Jeffrey A. Johnson Phone: (314) 232-2248 McDonnell Douglas Aerospace-East FAX : (314) 777-1328 P.O. Box 516 Email: m191713@svmstr07.mdc.com MC 1064126 St. Louis, MO 63166 ------------------------------------------------------------------------------- From: pereira@alta.research.att.com (Fernando Pereira) > Are you using PVM today? No. I tried it briefly a while ago, but I have not had the time to try it out yet for the project I have in mind. > What are your application(s)? Training statistical models of natural-language text. The training technique involves partitioning a large training set among processors, having each build improved parameter estimates from its training material, averaging the resulting parameter estimates and reiterating. There is a natural way of organizing this computation for PVM, and in fact Yves Schabes at Mitsubishi Research Labs in Cambridge (schabes@merl.com) with whom I have collaborated has used PVM in exactly this way. > How many and what kind of machines have you used with PVM? A few SGI R4000 Indigos . Fernando Pereira 2D-447, AT&T Bell Laboratories 600 Mountain Ave, PO Box 636 Murray Hill, NJ 07974-0636 pereira@research.att.com ------------------------------------------------------------------------------- From: fatoohi@nas.nasa.gov (Raad A. Fatoohi) Are you using PVM today? Yes What are your application(s)? NAS Parallel Benchmark (CFD applications). How many and what kind of machines have you used with PVM? About 40 SGI workstations. Thanks! Rod Fatoohi, Ph.D. Mail Stop T045-1 NASA Ames Research Center Moffett Field, California 94035 Phone: (415) 604-3486 FAX: (415) 604-3957 E-mail: fatoohi@nas.nasa.gov ------------------------------------------------------------------------------- From: wieland@bobafet.nrl.navy.mil (Fred Wieland) Here's an answer to some of your PVM questions on the survey. Are you using PVM today? I have a program which uses multiple processes on a single machine. I have used PVM to distribute the processes to multiple machines, and it worked successfully. I demoed the resulting product, and it received only lukewarm reaction from the community. I have since had little or no time to continue working on it, although it is definitely an active product as far as I'm concerned. What are your applications? The application area is physics-based modeling and simulation. How many and what kind of machines have you used with PVM? I have run PVM with the processes mentioned above running on an SGI Indigo, and SGI Crimson, and an SGI personal IRIS. I have used at most 3 machines at any one time, because I've only configured 3 of the 12 processes with PVM. I plan to configure all 12 sometime in the future, although I don't know when. I hope this information helps. -Fred Wieland, NRL code 7604, Washington, DC ------------------------------------------------------------------------------- From: Robert P. Goddard PVM Survey, 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: Walter Scott Are you using PVM today? Yes, we still use it, although I am not sure today that the choice to use PVM was a wise one. We have experience with both PVM 2.4 and PVM 3.0 patch 1-2. We are primarily interested in hooking up HOMOGENEOUS machines, and so the immense overhead of packing/unpacking isn't worth paying for. I have rewritten our communications so as to use sockets, and get much better performance. What are your application(s)? We ( Florian Mueller-Plathe and myself Walter Scott) have developed a parallel Molecular Dynamics code written in FORTRAN-77. Parallel MD on workstation clusters has been difficult to achieve -- indeed, I would say has not been efficiently achieved to date-- because of the relatively high communication/calculation ratio. The ration, btw, is MUCH better for quantum chemical problems, and correspondingly many quantum code have been successfully parallelized. We have come up with methods which we hope will aleviate the communication problem ( publications in preparation). Quite a few MPP vendors promise to offer a PVM port to their HW, and so, hopefully we would be able to use Iour code there as well. See below for a comment on this. How many and what kind of machines have you used with PVM? 1) up to 20 sun WS IPXs connected by Ethernet. PVM 2.4 worked well, PVM 3.0 is slower! 2) 3 IBM 6000 WS connected by Ethernet and SOCC. Used a highly unsatisfactory PVM version (called PVMe) for the SOCC connection: the pvm abstraction is not supported properly: the programmer has to watch out for dead locks( buffer allocation problems), bcast or mcast NOT supported,.... Also used the Ethernet connections with pvm 2.4. This worked, but of course the comm. bandwidth is tto small. 3) On a 64 processor KSR1. The PVM 3.0 version would not compile properly. The release was faulty in that the c routine identifiers were missing a "_". When I did get it going, performance was DISMAL. 4)port of our code to Intel Paragon by intel people. They replaced our PVM 2.4 communications with intel specific ones. IN comparing the the versions, the PVM version is much slower. From all of this I conclude: use PVM for portability ( but even then some vendors use PVM 2.4 and others use PVM 3.0 -- as far as I can see, the changes in user interface are cosmetic, provide few if any improvements and are simply a p. in the a.) BUT NOT FOR PERFORMANCE. This is true for WS clusters and even more so for MPPs. I have no experience using your graphic dooh-dahs, as I don't need them. Consider the following: the single most important reason for going parallel is the quest for performance. As far as I am concerned, PVM has yet to deliver. with kind regards, Walter Scott, Laboratory of Physical Chemistry, ETH Zuerich, Switzerland Tel. 0041 (1) 256 55 03 Fax. 0041 (1) 252 34 02 ------------------------------------------------------------------------------- From: Pedro A.M. Vazquez > Are you using PVM today? Yes, > What are your application(s)? a) Quantum chemistry calculations b) Molecular Graphics applications > How many and what kind of machines have you used with PVM? Versions 2.4.1 and 3.1 on AIX, SunOS4.1.1-2-3 and 386bsd0.1 (Jolitz) AIX: two 320h and 8 model 560 SunOS: a mix of 8 sparcstatios1+, ELCs , 330 and 390 servers 386bsd0.1: 386/387 40MHz pc clone, 486 33MHz pc clone > Any comments and suggestions for improvement are welcome. Sugestion: to create a mail exploder to redistribute comp.parallel.pvm to the people without news access. Now we have set up a listserver to do this in Brasil but it seems these people can't post through cs.utexas.edu to c.p.p. Sugestion: what about a pvm examples/contrib repository? -- ============================================================================= Pedro A. M. Vazquez | Caixa Postal 6154 | FONE: 55 192 39 7253 Depto de Fisico Quimica | CEP 13081 | FAX : 55 192 39 3805 Instituto de Quimica | SP/BRASIL | UNICAMP | |vazquez@iqm.unicamp.br ============================================================================= ------------------------------------------------------------------------------- From: zeitler@trantor.si.com (Dave Zeitler) > Are you using PVM today? No, but will be shortly. > What are your application(s)? Empirical tests of parallel random number generation using S-Plus running concurrently on a network of sun stations and/or an Ncube II. > How many and what kind of machines have you used with PVM? I've run up to 12 Sparc's at once before while working with PVM 2.4. The network I'll be working on now is 18+ Sparc's ranging from Sparc-1 to Sparc-10's. > Any comments and suggestions for improvement are welcome. > Please send the information to pvm-survey@cs.utk.edu. > Your help is much appreciated. Not right now, but I'll likely have some comments after I get the random number stuff running and start running the testing. -- Dave Zeitler zeitler@si.com Smith's Industries Ph: (616)241-8168 Aerospace & Defense Systems Inc. Fax: (616)241-7533 ------------------------------------------------------------------------------- From: Ludek Matyska > Are you using PVM today? I am not using PVM directly, but my PhD student (one) is working with HeNCe and PVM (and will definitevely work with after holidays). > > What are your application(s)? The computational chemistry, e.g., flexibility analysis (a paper about using PVM was sent to J. Comp. Chem. but, alas, I don't know its current state), molecular modelling (we would like to port Allinger's MM3 to parallel (distributed) environment using the PVM). > > How many and what kind of machines have you used with PVM? The largest VM used consisted of 8 Sparc's 1+ plus 4 DECStations 5000/120. We have some preliminary results using BSDI and SCO UNIX on Intel 80486 based boxes. Best luck with the PVM development Dr Ludek Matyska UVT MU Brno School of Informatics Buresova 20 City University 602 00 Brno London Czech Republic UK e-mail: ludek@muni.cz ------------------------------------------------------------------------------- From: wangy@gmsis.ms.ornl.gov (Yang Wang) > Are you using PVM today? Not today. But, will be using it tommorrow. > > What are your application(s)? ab initio electronic structure calculations. > > How many and what kind of machines have you used with PVM? 6 machines. DEC-stations, IBM-stations, iPSC-860, Paragon, etc... ------------------------------------------------------------------------------- From: Al Stone >Are you using PVM today? Yes, but on a sporadic basis. >What are your application(s)? I typically act more as a "help desk" than an end-user, due to the nature of my job. >How many and what kind of machines have you used with PVM? Up to 9 HP9000/735s on ethernet and/or FDDI; I have worked with others running 60+ machines (ranging from SPARC2's to other HPs to whatever they could get their hands on). More often than not, the usage has actually been on homogeneous clusters of machines. >Any comments and suggestions for improvement are welcome. >Please send the information to pvm-survey@cs.utk.edu. >Your help is much appreciated. Could you please indicate what some of the upcoming features might be? Or provide a mechanism for discussing them? Or, even provide a "beta" version? Thanks for asking! -- Ciao, al aka Al Stone Performance Consulting Team WSG Technical Consulting Workstation Systems Group (WSG) Hewlett-Packard Company 3404 E. Harmony Rd. Fort Collins, Colorado 80525 E-mail: al@hp_rigel.fc.hp.com Phone: 303-229-2399 Fax: 303-229-3002 ------------------------------------------------------------------------------- From: phanau@wv.MENTORG.COM (Paul Hanau) > Are you using PVM today? Yes. > What are your application(s)? Here at Mentor Graphics we develop EDA (Electronic Design Automation) software. I'm in a technology-oriented group responsible for shared "framework" technologies -- I work in the area of distributed computing support and intertool communications. I've been showing PVM to various end-user development groups to see if it might help them solve their problems in areas like distributed circuit simulation, etc., rather than re-inventing a similar system from scratch. > How many and what kind of machines have you used with PVM? I have PVM going on SUN4 and HPPA workstations thus far. I've been using the fractal display demo to illustrate the use of PVM, with up to 20 workstations involved, but 6-10 is more typical. > Any comments and suggestions for improvement are welcome. I've modified the demo to handle missing servers more gracefully and allow you to click on a band to identify which server handled it. This is useful in identifying slow-pokes (i.e. servers that take longer to do their piece than anyone else, in which case it might be faster to delete them from the configuration). If you'd like the modifications, let me know and I'll send them along. I've had no real problems with the software thus far. The documentation is very good (better than a lot of systems that cost money). If you have any references to uses of PVM in EDA I'd be interested in seeing them. Keep up the good work. ------------------------------------------------------------------------------- From: seyfarth@whale.st.usm.edu (Ray Seyfarth) > Are you using PVM today? Yes. > What are your application(s)? > Ocean modeling and parallel algorithms in general. > How many and what kind of machines have you used with PVM? > IBM RS6K 3 SGI 2 SUN4 3 ------------------------------------------------------------------------------- From: Dick Wilmoth > Are you using PVM today? Yes. > What are your application(s)? Direct Simulation Monte Carlo (flow simulations in rarefied gas dynamics) > How many and what kind of machines have you used with PVM? Sun Workstations (up to 32) SGI Workstation (4 processor) Cray YMP (8 processor) -Dick Wilmoth- ============================================================================ NASA Langley Research Center Email: wilmoth@ab00.larc.nasa.gov Mail Stop 366 Phone: (804) 864-4368 Hampton, VA 23681 ============================================================================ ------------------------------------------------------------------------------- From: sunny@cs.wm.edu (J. Sunny Egbo) |>Are you using PVM today? Yes. |>What are your application(s)? We are developing a set of C Library routines to facilitate running of Intel-like (Intel iPSC/2) commands within PVM. We are in the final stages of our testing. We are interested in running parallel simulations over PVM. So far everything is working like a charm except for the problems I have pointed out below. |>How many and what kind of machines have you used with PVM? About 11 IBM RS/6000 workstations. |>Any comments and suggestions for improvement are welcome. |>Please send the information to pvm-survey@cs.utk.edu. |>Your help is much appreciated. When we started using PVM, we experienced some sporadic hanging of the PVM daemons. We think that RS/6000 kills the largest process in the system if it runs out of paging space and hence PVM was getting killed. We have bumped up paging space on all our workstations and the problem seems to have gone away. Also, the timing of sends and receives over PVM is still unclear to us. A few months ago there were discussions on comp.parallel.pvm where someone posted a timing program (i believe it was called times3 or so). We ran this code between two of our workstations but were unable to get performance even close to what was reported by others. Any additional information about what performance numbers that we should expect given a network configuration will be appreciated. -Sunny ------------------------------------------------------------------------------- From: sfleisch@hydra.convex.com (Steve Fleischman) > Are you using PVM today? Yup. > > What are your application(s)? Molecular dynamics, neural networks. > > How many and what kind of machines have you used with PVM? HP clusters. > > Any comments and suggestions for improvement are welcome. > Please send the information to pvm-survey@cs.utk.edu. > Your help is much appreciated. > Fix the bugs in 3.1. Pvm_spawn, pvm_tasks, pvm_mytid. Also, please limit the anticipatory documentation. I.e., a feature is planned so the documentation reflects what the function will do when that feature is implemented not what is in the code currently. E.g., pvm_tasks and the where parameter. The "heuristic" used by pvm_spawn. The documentation should mention pvm_tidtohost. This routine is very useful. Otherwise, keep up the good work. ------------------------------------------------------------------------------- From: fsset@corelli.lerc.nasa.gov (Scott Townsend) Are you using PVM today? yes, 3.1 patchlevel 5 & 3.2mpp What are your application(s)? no real application yet, I'm primarily just trying to see what PVM does & how at this point. My current vehicle for these investigations are the NAS parallel benchmark codes BT, LU, and SP ported from the iPSC library versions. How many and what kind of machines have you used with PVM? Primarily 32 RS6000's in a cluster, also a 32-node iPSC/860 Any comments and suggestions for improvement are welcome. Warning: I used to be involved with the development of APPL, another message passing library. Most, if not all, comments below are basically derived from this biased viewpoint. (these are in no particular order) - My first surprise/struggle was with PVM's execution setup being quite different from what I expected in a number of aspects: - though I specify a directory for my remote executable, that executable's current directory is NOT set to the directory I specified, but rather my home directory on the remote machine. I can see why this comes about, but it's quite inconvenient. I'd much rather the pvmd put the default directory where the executable is found, at least if it's been specified. I know I could make the executable a script which does the cd, but then I have to specify the directory to both PVM and the script. - Interrupting a run with control-c isn't propagated to all processes. In general, I prefer any error exit by any process to bring down the other processes. (i.e. one process dumps core => other processes halt) I suppose this could be implemented with the notification stuff, but that's both a slight pain for the user and currently not possible since apparently the notification message doesn't carry the exit status of the process. - TIDs are a pain. Invariably I want to think in terms of process ID's runing from 0 to N-1. Thus I always have to map my ID to a TID. - The API requires too much work on the part of the user who's trying to get the optimal performance. For the simple & common case of sending an array of numbers, PVM should be able to make the decision about what encoding scheme to use, handle the initsend, pack, send stuff. I'm thankful for the low-level access to the alternatives, but most users I've seen don't want to hassle with all the alternatives and then either go PVMDEFAULT of PVMRAW everywhere, breaking the application in a heterogeneous environment or running unnecessarily slow. PvmDataInPlace would often help if it existed. I realize the problem for multicasting for doing optimal packing, but 90% of the time it's a send directed to a specific process. A higher level API could get optimal performance for 90% of the cases which are really quite simple. (simple things should be easy, complicated things should be possible) - There appears to be a significant performance hit on the iPSC. I haven't traced it down yet but I'm suspecting it's the XDR & buffer copying that's hurting us. Again, PvmDataInPlace, automagically selected by the PVM library, would be a real help. - There are a few applications here which find it useful to have multiple processes an a single RS6000. (multi-block CFD codes) PVM's performance within a processor, even using PvmDataRaw and PvmRouteDirect is poor. I'm getting about 2MB/sec for large messages and about 1msec latency for small messages. For comparison, I get 30 MB/sec thruput & 200 usec latency with APPL. The reason for this is the difference between the PVM socket/pipe vs. APPL's shared memory. I think it would be real useful if PVM used shared memory between processes within a workstation. From my APPL experience I know this isn't easy, especially the correct cleanup aspect, but the performance increase is significant. - This is probably my own error, but I haven't figured out what's wrong yet: our cluster of RS6000's has two seperate Ethernets along with an IBM ALLNODE switch. We can accesses these via IP names of lace01, lacemp01, and lacev701 respectively. It seems that I can only use the "native" lace01 name as the first name in the hostfile. Although I'm logged into lace01, which is the same machine as lacemp01, if the hostfile starts with lacemp01 I get various errors I don't understand quite yet. (lace01 is considered "native" in that that's what is returned by `hostname`) If there is some code internal to PVM which compares gethostname() with hostnames in the hostfile there could be a problem with machines supporting more than one network. - Online manual pages would be nice. With the AWF package you don't have to assume the site has DWB/nroff. - pvmd's idea of static load balancing is better than nothing, but doesn't allow for muti-user realities. For the cluster here we've built tools to grab a load average snapshot and assign processes on that basis. Not great, but better than assuming the machine is all one user's. APPL made building such tools straightforward through the concept of a process definition file. I've done something similar for my APPL over PVM library, using the SPMD example as a basis. It would be handy if either pvmd knew about ruptime/rstatd/etc or there was some way for a user to alter pvmd's concept of a host's load. (maybe this could be done with machine speeds somehow, I don't know) I'm trying to address some of the above issues for myself in APPL and iPSC compatibility libraries I'm building on top of PVM. This should do fine for me, but it's possible other PVM users out there have similar concerns with the current state of things. ------------------------------------------------------------------------------- From: sanjai@ctd.comsat.com (Sanjai Bhargava) pvm-survey> Are you using PVM today? No. But I have used it for projects in the past. ( About 6 months ago ). pvm-survey> What are your application(s)? I had used it in video coding, primarily to distribute the motion-estimation processing to different machines. pvm-survey> How many and what kind of machines have you used with PVM? I have used it on HP-9000 series machines ( 720's and 730's ), SGI Crimson VGXT and SUN-4. The maximum number of machines at any time were about 20. pvm-survey> Any comments and suggestions for improvement are welcome. Thanks for creating a useful application. The only drawback I felt were that there was no way to dynamically get the machines available. If any machine in the list was the down, the program would bomb. Also the graphical tools ( HENCE ) was not very useful and not well documented. Thanks, once again. sanjai -- Sanjai Bhargava COMSAT Labs. (301)-428-4502 ( phone ) 22300 Comsat Drive (301)-428-9287 ( FAX ) Clarksburg, MD 20871 e-mail: sanjai@ctd.comsat.com ------------------------------------------------------------------------------- From: Mohamed Faisal >Are you using PVM today? No. >What are your application(s)? The main work I did was to convert an existing program named SAP (By Dr. Raphael Finkel), which was originally written for Hypercube, Sun , Sequent etc , I rewrote it using PVM function. The goal was since PVM already takes care of the underlying architecture in SAP we dont need to worry about anything. The SAP which stands for Simple Application Package was written to help write parallel programs. >How many and what kind of machines have you used with PVM? Two. Sun workstations and Intel Hypercube. SUGGESTION: I felt the need for something like asyncronous receive (like intels hreceive) where you can pass a pointer to a function to the receive function and then do your own work. When there is any input , it should cause an interrupt and call the users function. ------------------------------------------------------------------------------- From: thomas@greisen.physics.utah.edu (Stanton Thomas) Are you using PVM today? Yes. What are your application(s)? I am using PVM to ray trace various optical detector configurations for use in monte carlo simulations of cosmic ray observatories. The ray tracing results are subsequently used via lookup tables in PVM implementations of the monte carlo. How many and what kind of machines have you used with PVM? I have used PVM on IBM RISC 6000 (RS6K), Decstation 3100 (PMAX), and on Silicon Graphics (SGI) machines. Any comments and suggestions for improvement are welcome. We have 14 Motorola 68030 based machines which are currently running the VxWorks operating system. VxWorks is a real time operating system with TCP/IP. I would like to be able to extend PVM to this system. We will have from 54 to 162 machines within the next three years. If anyone is working on a port for VxWorks we would be very interested. If not, do you have any notes/suggestions for developing PVM ports. Stan Thomas Cosmic Ray Research University of Utah (801) 581-8650 ------------------------------------------------------------------------------- From: kernst@jpmorgan.com (Ken Ernst) PVMers, I am currently using PVM 3.1. Genetic Algorithms and Monte Carlo simulation. Up to 20 nodes in a heterogeneous configuration of IBM RS/6000 and SUN SPARC2s, IPXs, SPARC10 model 41s. Have the pvm_config command return the correct number of architectures and indicate a model number. For example SUN4 IPX. And give us a way to set the SPEED parameter. Regards Ken P.S. Great product. ------------------------------------------------------------------------------- From: delbaeth@mule.fhcrc.org (Robert Christ) >Are you using PVM today? We currently have PVM running but it is not being used. We have plans to use it in the future but due to some problems in code we have tried to switch to PVM things are progressing slowly. >What are your application(s)? > >How many and what kind of machines have you used with PVM? Right now it is running on 5 RS6000s. We plan to include a large pool of SPARCstations in the near future. Rob Christ Unix Network Administrator Public Health Sciences Division Fred Hutchinson Cancer Research Center ------------------------------------------------------------------------------- From: roscoe@BUCRF15.BU.EDU (Roscoe Giles) From: Roscoe Giles Deputy Director, Center for Computational Science Boston University Boston MA, 02215 617-353-7082 roscoe@roscoe.bu.edu ARE YOU USING PVM TODAY? yes WHAT ARE YOUR APPLICATION(S)? (1) We have a newly developed set of courses on parallel scientific computing with a special emphasis on MPP. PVM is being used to develop example applications linking workstations to our CM-5. There is a little work using PVM as a model distributed parallel system. (2) Some of our researchers are using PVM for research applications similar to the teaching application (1) -- ie human interface and graphics on workstation/ linked to CM-5. (3) One group has used PVM and hence to link a workstation cluster on campus to one at a local High School which is linked to us. HOW MANY AND WHAT KIND OF MACHINES HAVE YOU USED WITH PVM? CM-5 (here, NCSA, TMC) SGI SUN RS6000 ------------------------------------------------------------------------------- From: Jari Toivanen > Are you using PVM today? Not at the moment. > What are your application(s)? Solving partial differential equations. > How many and what kind of machines have you used with PVM? Few HP and Sun workstations. Jari Toivanen (toivanen@jyu.fi) Memories are lost in time University of Jyvaskyla, Finland like tears in rain. ------------------------------------------------------------------------------- From: tadams@bach.NOSC.Mil (Thomas A. Adams) > Are you using PVM today? Yes. > What are your application(s)? I have implemented my own "standardized" general purpose master-slave computing software (broadcast, parallel computation, commutative-associative collapse repeated over and over). > How many and what kind of machines have you used with PVM? Primarily Sparcstation networks. I have done a little bit with Convex. Attempted to use SGI but without success. I expect to use the Paragon in a few months. > Any comments and suggestions for improvement are welcome. The user's guide (ORNL/TM-12187) is good, but it might be worthwhile to also have (perhaps in a separate document) more examples and guidelines for good PVM programming practice. A standard PVM test suite would be good, if you do not already have one. ------------------------------------------------------------------------------- From: swk@mlb.semi.harris.com (Song Koh) pvm> Are you using PVM today? Yes. pvm> What are your application(s)? General program to program communication. In the domain of semiconductor device modeling. Communication between C programs, Scheme Programs, Spice simulators, Splus statistics package plus a few others. pvm> How many and what kind of machines have you used with PVM? Mostly SUN SPARC. But also HP9000, and DEC MIPS. pvm> Any comments and suggestions for improvement are welcome. Good job! ------------------------------------------------------------------------ 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: lucier@math.purdue.edu (Brad Lucier) > > Are you using PVM today? Not yet. > > What are your application(s)? > Flows in porous media > How many and what kind of machines have you used with PVM? > > Any comments and suggestions for improvement are welcome. > Please send the information to pvm-survey@cs.utk.edu. > Your help is much appreciated. > > Best wishes, > The PVM Project Team > > Brad Lucier ------------------------------------------------------------------------------- From: "Dr. Eric C. Frey" > > Are you using PVM today? we are in the process of developing code that uses it. We don't have any production applications running yet that use it (a few test applications). > > What are your application(s)? > We are writing applications that do 3-D reconstruction of single photon emission computed tomography images including compensation for physical image degrading factors. This is essentially a matrix inversion problem on very large matrices (8192x8192) using very noisey data. > How many and what kind of machines have you used with PVM? > we have 8 decstations (5-5000/200's, 1-2100, 1-3100); 1 stardent 3020, 2 stardent 750 (3020 in small box). We are planning on buying 2 Alpha 400's. > Any comments and suggestions for improvement are welcome. > Please send the information to pvm-survey@cs.utk.edu. > Your help is much appreciated. > Our application could benefit from use of multicasts to distribute data sets. Also, help in load balancing (or providing the information to do it) and/or fault recovery. Eric Frey ------------------------------------------------------------------------------- From: harkin@fubar.cs.montana.edu (Gary Harkin) } } Are you using PVM today? Yes. } } What are your application(s)? Distributed systems for performance improvement. Primarily statistical mechanics. } } How many and what kind of machines have you used with PVM? Mips-based Decstations, HP 9000's. ------------------------------------------------------------------------------- From: solberg@radonc.ucla.edu (Tim_Solberg) We are using PVM to improve the performance of Monte Carlo radiation transport calculations for medical physics applications. We are running the MCNP Monte Carlo code developed at Los Alamos NAtional Labooratory. Our present version uses PVM 2.4, but I understand that a version using PVM 3.1 is ready for beta testing. We should begin this here in the next 1 to 2 weeks. We currently run on several Suns (SPARC2s) within our department and several more in other departments on campus. We have two MP workstations (4 SPARC2 processors each), thus PVM 3.1 should ofer substantial performacnce improvement once the Sun MP platform is supported. MCNP/PVM also runs on RS/6000s. For more information on MCNP/PVM, contact Gregg McKinney (gwm@lanl.gov) at Los Alamos. Thank you for your continued efforts on PVM. Timothy D. Solberg UCLA School of Medicine Dept. of Radiation Oncology 200 UCLA Medical Plaza, #B265 Los Angeles, CA 90024-6951 solberg@radonc.ucla.edu 310-825-4323 ------------------------------------------------------------------------------- From: Alan Ferrenberg > > Are you using PVM today? If you mean "have I run PVM today, Monday August 16?" the answer is no. I've used it within the past week. > > What are your application(s)? Molecular Dynamics simulations of fiber growth Cluster-Hybrid Monte Carlo simulation algorithms > > How many and what kind of machines have you used with PVM? RISC/6000s (several configurations with up to 7 processors) 386 PCs running BSD/386 (2 PCs) up to 15 processors on an IBM SP1 ------------------------------------------------------------------------------- From: sverre@nrel.gov (Sverre Froyen) > > Are you using PVM today? Yes. > What are your application(s)? Electronic structure calculations for solids. > How many and what kind of machines have you used with PVM? Sun Sparcstations, IBM RS6000. > Any comments and suggestions for improvement are welcome. A global syncronization call (pauses until all processes have caught-up). -- Sverre Froyen (+1-303-231-1782; FAX: +1-303-231-1271) NREL (National Renewable Energy Laboratory), Golden, CO 80401 email: sverre@nrel.gov ------------------------------------------------------------------------------- From: dwhite@mrj.com (David White) !. 1. Yes, I'musing pvm. 2. It serves as the basis for a mesage 2. It serves as the message passing system for a distributed application. 3. We are using Sparcs and a CM5. Dave (dwhite@mrj.com) ------------------------------------------------------------------------------- From: mobarry@maxwell.gsfc.nasa.gov (Clark M Mobarry) 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? SGI Indigo's, MasPar front end (Decstation), YMP-EL ------------------------------------------------------------------ Clark Mobarry mobarry@maxwell.gsfc.nasa.gov NCCS | voice: (301) 286-2081 Code 934, NASA/Goddard Space Flight Center | FAX: (301) 286-1634 ------------------------------------------------------------------ ------------------------------------------------------------------------------- From: fred@network.ucsd.edu (Fred Criscuolo) Hi, Yes we are currently using PVM. Our primary apllication right now is to do concurrent searches of the Brookhaven Protein Database on a series of complex keyword searches. We are also hoping to develop a PVM version of a critical component of a computational chemistry code. We are using SGIs and RS/6000s. Fred fcriscuol@ucsd.edu ------------------------------------------------------------------------------- From: js9s@uvahey.phys.Virginia.EDU >>Are you using PVM today? No, but I used it yesterday. >>What are your application(s)? I use it for data processing. Basically tape io speed is about 6 times the proceessing program speed, so it's natrual try to use one tape io device and distributing the events to other machines. >>How many and what kind of machines have you used with PVM? I run it on DECstations and SGI. about 15 of them. >>Any comments and suggestions for improvement are welcome. I know some fellow outside the PVM project is trying to port the thing to VMS. That's a very useful stuff considering the huge left-overs from last 15 years. thanks .j.g. ------------------------------------------------------------------------------- From: rsimms@math.clemson.edu (Robert Simms) Just recently used PVM in my masters project. Implemented a linear (least squares) solver using Givens rotations to obtain a triangular form of a large matrix. Givens transformations were done in parallel using PVM. Large number of communications seemed to hang daemons, so I bundled transformations together in messages to cut down on number of messages sent, but still, if I use a large enough matrix the daemons time out trying to communicate with each other. Tried to ask around for help, but only got: "pvmd could get swamped with messages" I used Sparcstations running SunOS 4. Roughly 4 of them were at my disposal. - Rob ------------------------------------------------------------------------------- From: fshyun@borey.lerc.nasa.gov (Hyun Kim) > > 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? IBM RISC machines ------------------------------------------------------------------------------- From: Po Shan Cheah > > Are you using PVM today? Yes. > > What are your application(s)? Multivariate integration in a distributed processing environment. > How many and what kind of machines have you used with PVM? Only 4 Sun Sparcstations. +----------- | Po Shan Cheah Columbia University SEAS | | cheah@cs.columbia.edu New York, NY | pcheah@nyx.cs.du.edu -----------+ ------------------------------------------------------------------------------- From: duncan@mercury.chem.utah.edu (Thomas W. Duncan) At this moment, I am working on upgrading some of our software from pvm2.4 to pvm3.1.5. We are using it to solve transtition state theory problems in theoretical chemistry. Up to this point, we have been running all of our code on a cluster of RS6000's (4), but will probably be adding an SGI machines to this soon. I would just like to say that compilation of the source code on the RS6000's was VERY easy. The makefiles worked perfectly and made me very happy. Wendell Duncan Chemistry Dept. University of Utah ------------------------------------------------------------------------------- From: vieux@chief.ecn.uoknor.edu I have compiled pvm on an IBM RS6000 Model 520 but have not really used it yet. When running it would be on 2-RS6000 Model 220's and the 520. Baxter Vieux ------------------------------------------------------------------------------- From: Peter McGavin >Are you using PVM today? So far I've only run some of the examples. >What are your application(s)? Intend to parallelise existing meso-scale wind model (meteorological). >How many and what kind of machines have you used with PVM? 5 Sun SPARCstations (SUN4) - 2xSPARCstation 1, SPARCstation 2, ELC, SLC 2 HP workstations (HPPA) - HP9000/705, HP9000/730 1 DECserver 5400 (PMAX) - 5400 Regards, Peter McGavin. ------------------------------------------------------------------------------- From: Peter Geschiere > > Are you using PVM today? I started with some experiments. > What are your application(s)? A parallel Navier-Stokes solver. > > How many and what kind of machines have you used with PVM? A 4 processor IBM RISC System/6000. > > Any comments and suggestions for improvement are welcome. > Please send the information to pvm-survey@cs.utk.edu. > Your help is much appreciated. I am planning to use pvm version 2.4. I know that version 3 has more primitives and is supplied with a much better manual, but I also heard that this version still contains some bugs. Is this true, and if so, when do you expect to have these bugs solved? Regards, Peter Geschiere. ------------------------------------------------------------------------------- From: "Andrew J. Caird" Are you using PVM today? No, I am just learning it. What are your application(s)? Hopefully paralleization of neutron diffusion codes used in nuclear reactor analysis. How many and what kind of machines have you used with PVM? Have tried with IBM RS6k and HP 700 series. As above, I am a true beginner and thus have really nothing to show for it. Thank you. --Andy ------------------------------------------------------------------------------- From:davis@masig.fsu.edu IRIS workstations and our CRAY YMP. >Any comments and suggestions for improvement are welcome. The computing center has not upgraded to version 3, whereas I have done this for the application that runs on the workstations. The restriction that both server and client must run the same version is restrictive. It may be desireable for major changes, 2.4 to 3.0, but should be avoided for minor ones like 2.4.1 to 2.4.2. I have several workstations and many users on each and I would like to install a single version of pvm for all machines/users. This is rather difficult for the current version. >Please send the information to pvm-survey@cs.utk.edu. >Your help is much appreciated. > >Best wishes, >The PVM Project Team > _______________________________________________________________________________ Alan Davis | INTERNET davis@masig.fsu.edu Mesoscale Air-Sea Interaction Group | The Florida State University | Room 8A Love Bldg. | Phone 904-644-3798 Tallahassee, FL 32306-3041 | FAX 904-644-4841 _______________________________________________________________________________ ------------------------------------------------------------------------------- From: kern@masc5.rice.edu (Michel Kern) > > Are you using PVM today? Not today, Aug. 16th, but this week, probably yes. > > What are your application(s)? In a broad sense, seismic inversion. More precisely, we (repeatedly) solve PDE's (for instance, the wave equation) with multiple right-hand sides. We use PVM to distribute the RHS over different computers. > > How many and what kind of machines have you used with PVM? Not more than 8. Mostly SUN's (Sparc 1, 2, 10), IBM RS6000, and soon HP. Tried on Paragon, no luck. > > Any comments and suggestions for improvement are welcome. Thanks for the very useful tool, and the support. -- Michel Kern Department of Computational and Applied Mathematics Rice University, PO Box 1892, Houston Texas 77251-1892 Tel: (713) 527 8750 x2288 E-mail: kern@rice.edu ------------------------------------------------------------------------------- From: Andrew Lumsdaine Are you using PVM today? Yes. What are your application(s)? Semiconductor device simulation. Waveform methods. How many and what kind of machines have you used with PVM? o Cluster of eight RS/6000 workstations with Sparc-2 as host o Cluster of 14 Sparc-10 workstations Any comments and suggestions for improvement are welcome. I recently read a tech report by Grant and Skjellum entitled "The PVM Systems: An In-Depth Analysis and Documenting Study." (I assume the PVM Project Team is aware of this report as several members of the team were listed in the acknowledgments). There seemed to be a fairly long list of cons and some of them seemed to be quite serious. Will these be addressed future releases of PVM? -------------------------------------------------------------------- Andrew Lumsdaine Dept. Comp. Sci. & Engr. email: Andrew.Lumsdaine@nd.edu 353 Fitzpatrick Hall phone: (219) 631-8716 University of Notre Dame fax: (219) 631-8007 Notre Dame, IN 46556 -------------------------------------------------------------------- ------------------------------------------------------------------------------- From: Oleg Svintsitski I am using PVM 2.x for parallel computing of the multistage stochastic programming problems. The idea is to split an original decision tree between machines by assigning its leaves to different processors with providing the information exchange between the root and leaves through PVM. I've been using up to 8 IBM RS/6000. The limitation of using IBM's only is caused by the availability of the OSL software that is used as a LP-solver in the algorithm. It seems that PVM (i.e. presence of communication between different machines) slows things down a lot. My research is intended to show the advantage of parallel computing for this kind of computations, but I am not getting too much advantage against just serial computing so far. Hopefully switching to PVM3.1 will give more flexibility and speed up the computations. If you have any suggestions about how to increase the communication speed, I would really appreciate them. Oleg Svintsitski. ------------------------------------------------------------------------------- From: skumar@cadcam.eng.ohio-state.edu (Sanjay Kumar) > Are you using PVM today? Yes. Very much. I am a grduate student and PVM is very forms the very core of my MS thesis related research. Well, its not PVM rather applications needing PVM. > What are your application(s)? Engineering applications : finite element analysis which reduces to large-scale equation solving and engineering design using genetic algorithm. > How many and what kind of machines have you used with PVM? 11 IBM RS/6000, CRAY, CM-2 (ethernet connection) I believe its a great effort. We are really fortunate to have such a portable system available in public domain. My suggestions: In the long run: I. There should be some way of visualizing the program performance. I am looking for something that can give us kiviat diagram and Feynman's diagram as seen in papers by Geist and Sunderam. II. Portable and integrated port to multiprocessors like CM-5 which will allow us to bypass the complicated CMMD and other such vendor-written message-passing systems (at least for simple tasks). That is, I would expect same program to run on say workstation cluster and CM-5. I know that's asking for too much but I am glad there are already some efforts on this direction. Short term: I Direct TCP/IP link between tasks. II PvmDataInPlace We should be thankful to Pvm_Team for giving us a good tool. Sanjay Kumar Graduate Student, Civil Engg. The Ohio State University E-mail : skumar@cad1.eng.ohio-state.edu ------------------------------------------------------------------------------- From: subba@lsil.com (Subba-Rao Kalari) > Are you using PVM today? We received PVM a couple of weeks back. We have not put it to use yet. > What are your application(s)? Our initial use will be understand the cost/benefit trade-off of distributing applications across the network. > How many and what kind of machines have you used with PVM? We will be using a heterogenous network of Sun and HP workstations. Thanks for making the results of your work available. Subba Rao ------------------------------------------------------------------------------- From: Elizabeth Yip Are you using PVM today? yes What are your application(s)? electromagnetics How many and what kind of machines have you used with PVM? iPSC/860 iPSC/2 Cray Sun Any comments and suggestions for improvement are welcome. How about an I/O package e.g. cread cwrite .... in the iPSC's ??? ------------------------------------------------------------------------------- From: csfb1!cowboys2!nstrauss@uunet.UU.NET (Neil Strauss) > > Are you using PVM today? We are currently evaluating PVM as an alternative to a proprietary tool` for parallelizing complex financial calculations related to Mortgage Backed Securities. There are several other tools under evaluation for the same purpose including Mentat, Linda, and various Symmetric MultiProcessor UNIX machines. > > What are your application(s)? see above > > How many and what kind of machines have you used with PVM? We have a network of several hundred SUN IPXes, and we have run parallel computations across 150 machines simultaneously using our proprietary coarse grained system. We have tested some calculations across a smaller network of 25 SUN IPCs and IPXs. ------------------------------------------------------------------------------- From: wiener@haze.rap.ucar.EDU (Gerry Wiener) Are you using PVM today? Not currently. What are your application(s)? Meteorological How many and what kind of machines have you used with PVM? None Thanks for your efforts. Gerry Wiener Email:gerry@ncar.ucar.edu ------------------------------------------------------------------------------- From: rwb@c3.crd.ge.com (Robert W. Benway) Are you using PVM today? *But-of-COURSE* What are your application(s)? Electromagnetics, fluid mechanics, medical diagnostic imaging, & others How many and what kind of machines have you used with PVM? 100-200 different systems. Suns, DECstations, HP-700s, Convex, CRAYs, SGI Any comments and suggestions for improvement are welcome. Comments (& brainstorms) have been given to ORNL/UTK as they occur.... -RWB ------------------------------------------------------------------------------- From: huangsl@stnick.me.vt.edu (100G Randolph) I am trying to use PVM today. My application is hydrodynamic instability ( it can be reduced to an algebraic eigenvalue problem ). I can access 9 RS6000. I just begin to use PVM, so I have no comment at this time. ------------------------------------------------------------------------------- From: ychung@fcusqnt.fcu.edu.tw (Dr. CHUNG Yeh-Ching) Yes, i am using PVM. Finite element computing. a cluster of workstations (4 - 16). Suggestion: 1. I think the communication primitives need to be enhanced, i.e., provides more communication functions. 2. It seems to me that everyone wants to run PVM must install PVM in their directories. Otherwise, you cannot execute pvm daemon. 3. The interface between users and pvm can be enhanced. Yeh-Ching ------------------------------------------------------------------------------- From: I am using pvm3.1 today I have 3 SUN4's, 1 HP-370 and 1 SGI/Indigo I am dividing up a finite difference algorithm, it must do a SEND/RECEIVE to each neighbor processor upon each entry into the main loop. I don't like the way the master processor will choose itself as a subprocess as a slave, this seems to defeat the purpose of pvm, I could have just caspawned a pro a process, or called a subroutine to do the same. I seem to have problems if i call a SEND to another processor which has not yet made it to RECEIVE, my program takes a long time to run. If I let the RECEIVE processor be the HP, which is much faster, then it works fine, since it is already to accept the SEND processors data. Is their a reason for this ? Are their ways to make each processor stay in sync with each other ? I am using the fortran commands, I only have a manual for 3.0, did 3.1 add many toe others ? ------------------------------------------------------------------------------- From: Armaghan Hoveydai > Are you using PVM today? > Yes > What are your application(s)? > Teaching parallel programming concepts to undergradute students. > How many and what kind of machines have you used with PVM? > 10-15 SUN SPARCstations Regards, +============================================================================+ | Armaghan Ross Hoveydai A.Hoveydai@cit.gu.edu.au| | Manager Computing Support Ph +61 7 875 5027 | | Faculty of Science and Technology Fax +61 7 875 5425 | | Griffith University, | | Nathan, QLD, 4111, | | Australia. | +============================================================================+ ------------------------------------------------------------------------------- From: Michael Rezny Yes I am using PVM today, in fact regularly. My applications are research into distributed processing and in particular modelling drought conditions in Australia for the Department of Primary Industries. We have used PVM on SUN4 and SGI, I have also used it on DEC and HPas well I have ported it for use with the MasPar and LINUX ( a unix clone for 386/486 IBM PCs) Mike REZNY ------------------------------------------------------------------------------- From: nhan@tiny.me.su.oz.au (Nhan Phan-Thien) 1. Yes, I'm current using PVM3.1 2. The application is three-dimensional Stokes flows past a large number of particles be an indirect boundary element method (called completed double layer boundary element integral method). The problem is large in scale (up to 200000 3D boundary elements). 3. I'm using PVM on a network of o up to 4 DEC alpha o two Titan Ardent o up to 6 SGI Indigo o up to 22 DEC 5000/200 The system works fine, but communication is sometimes a problem. We are currently write up a paper on the application of the method with PVM. Thanks you again for a nicely developed software. Nhan Phan-Thien, Professor of Mechanical Engineering. ------------------------------------------------------------------------------- From: glennw@glenrowan.acci.COM.AU (Glenn Wightwick) > Are you using PVM today? Yes, I am using PVM 2.4.2 and PVM 3.1 > > What are your application(s)? Applications are in meteorology and the environment and sorting. > > How many and what kind of machines have you used with PVM? I am using PVM with a cluster of 16 RISC System/6000 Model 350s and 1 RISC System/6000 Model 52H. ------------------------------------------------------------------------------- From: kinoshi@crayamid.cray.com (Toshihiro Kinoshita) > > Are you using PVM today? > Yes, I am. > What are your application(s)? > QCD (Quantum Chromodynamics) > How many and what kind of machines have you used with PVM? > CRAY Y-MP SGI IRIS 4D Regards, - Toshihiro Kinoshita (kinoshi@cray.com) ------------------------------------------------------------------------------- From: nakano@scs.ased.mt.nec.co.jp (Eiichi Nakano) >>Are you using PVM today? No, but I will try soon. >> >>What are your application(s)? Mathmatics library. >> >>How many and what kind of machines have you used with PVM? Three or four. Machine EWS4800,UP4800 (This machine is made by NEC cooporation. ------------------------------------------------------------------------------- From: obayashi@vdrl.src.mei.co.jp (Yoshimasa Obayashi) l Are you using PVM today? No. But I usually use PVM3.1 two days per week. l What are your application(s)? Mainly, Matrix Computations ( Eigenvalue Problem, Solving Linear Systems, FFT etc ). I am planning for using Fluid dynamics and FEM now. l How many and what kind of machines have you used with PVM? SUN4. l Any comments and suggestions for improvement are welcome. l Please send the information to pvm-survey@cs.utk.edu. l Your help is much appreciated. If I find the bugs or the errors, may I inform it to the same E-mail address ? Yoshimasa Obayashi VLSI Devices Research Laboratory Semiconductor Research Center Matsushita Electric Industrial Co., Ltd. E-mail:obayashi@vdrl.src.mei.co.jp ------------------------------------------------------------------------------- From: Sergey Ten Dear collegues, We are sending you the answers on your questions about our PVM interests: - yes, we are using PVM; - we are using PVM both in research and educational processes; - today PVM system is installed on the part of local network of University of Aizu. This part includes four SUN-4 SPARC stations. May be in future it will be possible to spread PVM system widely in this local network (it depends from and needs the special permission). P.S. Would you be so kind to inform us about the easiest way to obtain information about: 1. PVM-group (we have heard about its existance); 2. Have this group special pool (available through ftp, for instance) of PVM applications; 3. Does there exist PVM news-list and how can we obtain it. Thank you in advance, Sincerely yours, S.ergey Ten ------------------------------------------------------------------------------- From: Koichi Konishi |> Are you using PVM today? Not as of today. I used it for a few experimental applications. |> What are your application(s)? Work scheduler using genetic algorithm, trajectory calculation of mass points, ... |> How many and what kind of machines have you used with PVM? NEC EWS4800 workstations and SPARCstations. Regards, Koichi Konishi Computer System Lab., C&C Labs. Phone: +81-44-856-2178, FAX: +81-44-856-2231 ------------------------------------------------------------------------------- From: Brian J. Keefe >Are you using PVM today? We are attempting to centrally administer PVM and HeNCE on a network of 300+ workstation or x-terminals. I have prototyped several applications with it. Others have done more extensive number crunching. >What are your application(s)? I personally used it to develop threaded PVM nodes with Concurrent C and the C++ tasking library for prototype purposes. I tested lwp with RPC but could have and should have tried it with PVM. The goal was/is to maintain a single PVM which varied applications utilize as a network communications substrate, as an alternative to RPC or socket programming. PVM seems to potentially provide a happy medium between the RPC and sockets paradigms. >How many and what kind of machines have you used with PVM? Our network is currently homogeneous. I've prototyped my applications on 2 or 3 machines only. Others in the organization use PVM for production applications on, I believe, more machines. >Any comments and suggestions for improvement are welcome. >Please send the information to pvm-survey@cs.utk.edu. >Your help is much appreciated. - Allow a PVM to dynamically accept registrations from different users with some nominal security - Explicitly support threads within PVM processes - Allow PVM daemons to initiate joining an existing PVM using some well-defined PVM id of some kind, possibly using DNS services to in some way identify existing PVMs. - Support a PVMinetd to start remote PVM daemons -- Brian Keefe m1bjk00@fed.frb.gov ------------------------------------------------------------------------------- From: Yogesh S. Wagle > Are you using PVM today? Yeah, but on an experimental basis. > > What are your application(s)? Currnetly I am running the SIMPLE benchmark. I will be also running another fluid dynamics application soon. > > How many and what kind of machines have you used with PVM? 16. all SUN4s > > Any comments and suggestions for improvement are welcome. Improve the process-processor mapping algorithm. It might be possible to monitor the load on the machines before spawning the processes onto them. -- Yogesh Wagle ------------------------------------------------------------------------------- From: ricardo@ing.puc.cl (Ricardo Schmutzer von Oldershausen) > > Are you using PVM today? Yes. > > What are your application(s)? I am using PVM to parallize an optimization engine (SIMPLEX method) as my thesis. > > How many and what kind of machines have you used with PVM? DEC 5000s, a SparcStation 1, a SUN Workstation, a PC-386 (DTK) (up to 8 machines simultaneously). Ultrix 4.3.0, Ultrix 4.2.0, Ultrix 4.1.0, SunOS 4.1.2, Linux 0.99p9 > > Any comments and suggestions for improvement are welcome. > Please send the information to pvm-survey@cs.utk.edu. > Your help is much appreciated. Thank you for making PVM, it has helped me a lot with my thesis ! I ported PVM to Linux (took the BSD Makefile and had to modify just a few lines in 2-3 files, so that it compiled correctly. I have used it until now only in a stand-alone machine (at home ;-), but will propably port it to Linux 0.99p12 in two-three weeks in a networked configuration (together with the above machines). If you are interested in the port, please let me know, to send you the corresponding patch. Wishes ? Perhaps, a function to join two send buffers, let me explain why: I would like to be able to modularize software written under PVM, for example for writing libraries for PVM. These libraries should be able to receive send buffers as arguments and be able to include it's own data in front of the "user" data. Let's suppose you have a library of functions that uses PVM (in my case a pool of workers). These functions receive in a send buffer the data that must be sent to another process, but the library itself must pass along additional data AT THE BEGINNING of the message, so that the library stub at the receiver side can decode it first. At the moment, I must pass the user data in a structure and a user function that packs the structure, so that the library can make a send buffer, pack it own data and then call the user function to pack the user data. This arrangement is ineficient, because an additional copy of the data is required (and it would be nicer if you could just pass a send buffer descriptor to PVM libraries). Sometimes inside the library I recognize that there are many small buffers (is there a way to check the size of a buffer BEFORE sending it ?) for the same destination. Here also, it would be nice to be able to concanate all the buffers to send them in just one message (i.e. to send it as one big TCP packet, instead of sending many small ones. I suppose the PVM library itself does not make such an optimization (yet? ;-). So the user should be able to join buffers (of course this could be made when the buffers are made, but in my case when I pack the buffers I do not know yet to whom I will send them (this is resolved in a lower level where the load of each host is known, i.e. inside the library). Something like: int pvm_pkbuf(int bufid) Packs into the active send buffer all data members from the buffer bufid. Eventually, if it necessary for efficiency, the buffer bufid may be destroyed in the process (as if pvm_freebuf would have been called). Perhaps, in this way it is just necessary to change some pointers to make the operation (I haven't seen your code in detail ! I may be talking nonesense !!). Thank you ! Ricardo -- +----------------------+---------------------------------+--------------+ | Ricardo A. Schmutzer | Internet: ricardo@ing.puc.cl | Conectividad | | PUC, Santiago, Chile | | Distr. Comp. | +----------------------+---------------------------------+--------------+ ------------------------------------------------------------------------------- From: jjacob@paxvax.EE.CORNELL.EDU (Joseph Jacob) >Are you using PVM today? > Yes. >What are your application(s)? > I am experimenting with a particle simulation as a case study for our dynamic load balancing scheme for distributed computers. >How many and what kind of machines have you used with PVM? I've been running my simulations on up to 16 Sun Sparc 1+ workstations (PVM_ARCH = SUN4). > >Any comments and suggestions for improvement are welcome. Notes: 1) I found that my application can adversely affect other machines on the same network. One of my colleagues has reported that his workstation hangs and he gets the message "Ethernet jammed" when I run my program. Others have reported that I have no effect on them, even when their machine is in my PVM configuration. This may be due to the relatively heavy communication required in my particle simulation, rather than some problem with PVM. 2) I have been unable to get PVM to work for the following: a) two or more communicating processes on the same Sparcstation. b) ANY configuration of HP9000 700 series workstations (PVM_ARCH = HPPA) 3) The PVM documentation I have states that I should get a message like "PVM ready" when PVM is started with "pvmd3 hosts" (hosts is the host file). Then only message I get is something like "Tid ready", where Tid is the task ID of the process spawned on the same machine as pvmd3. I then have to keep checking the pvm console until the full configuration is in place. Thank-you. -- Joe Jacob ------------------------------------------------------------------------------- From: Toyama Sakuji >>Are you using PVM today? No >>What are your application(s)? Fasta >>How many and what kind of machines have you used with PVM? 5 Sparcstations, Sunos 4.1.2 Toyama Sakuji Institute for Virus Research Email stoyama@virus.kyoto-u.ac.jp ------------------------------------------------------------------------------- From: dunkel@carbon.chem.utah.edu (Reinhard Dunkel) I used PVM 2.? during the development of my analysis software for 2D INADEQUATE NMR spectra. A typical run was 70 RS/6000 CPU hours and I used my access rights to the physics student lab (20 RS/6000) to finish a run over night. I encountered numerous problems during with PVM 2.? (1) Starting pvmd required that all enumerated computers were really accessible. Pvmd did not recover from an unavailable machine and so I had to try several times (5 minutes each) to start pvmd until all currently "bad" computers were removed and pvmd would start. (2) Using a few local computers worked fine with PVM. When using a larger number of computers (>10) or remote computers I always ran into network problems. These problems were not caused by the average load that my application produced. Anyhow, PVM complained so much that it was impossible to determine whether or not any real progress was made. (3) Permission problems. I never figured out under what conditions computers could be used with pvmd. I am talking exclusively about RS/6000s here. To specify passwords from the .netrc file worked more reliably than to depend on any rsh permissions - so that's what I used. But lots of RS/6000s would not cooperate. I suspect the "secure network" option used on some RS/6000s was the problem but I never really figured it out. I did not find a section in the PVM documentation that explained the communication of the pvmd daemons and how the computers have to be administered to allow it to work. (4) In PVM 2.? I did not find a smooth way to tell from my program whether pvmd was running or not. Enrolling in PVM should fail with a "problem" return code if pmvd is not running rather than for PVM to print lots of error message and to die. So I had to modify the source code substantially (don't print error messages and time out quickly) to make it work. (5) When PVM 3.0 came along (February) lots of things which worked in 2.? did not in 3.0 and lots of stuff described in the new manual did not work either. I reported all problems I found and Robert Manchek (manchek@cs.utk.edu) confirmed that these were indeed problems. I didn't try a newer PVM version ever since. I improved the speed of my software so one run can be finished over night on one RS/6000. I found the over-night-run to be more practical than to use PVM. So PVM is supported in our software but nobody uses it regularly as far as I can tell. As datasets get bigger and fancier approaches to the analysis are introduced the run times of the analysis go up. If analysis times go over 10 CPU hours again I am sure PVM will get more attractive and I hope I'll discover that PVM got less painful to handle in the newer versions available now. __ _ | ) o | | | \ | | | / _ |_ _ _ _| | | | _ | |/\ /_) | ^ ^ | | _) | / | | | | | ^ ^ |_) /_) | | \_/\___|_| | |_| |_(_|__|___\_| |_/__|_|_| | |_| \_/\___| Department of Chemistry Office: (801) 581-7351 2020 Henry Eyring Building b113 Home: (801) 582-7516 University of Utah FAX: (801) 581-8433 Salt Lake City, UT 84112 dunkel@chemistry.chem.utah.edu ------------------------------------------------------------------------------- From: bessey@milo.math.scarolina.edu (Al Bessey) Are you using PVM today? Yes What are your application(s)? I am using pvm to write a parallel version of a Lagrangian Particle Dispersion Model which is a part of a visualization package for the Regional Atmospheric Modelling System (RAMS) code. I intend to also use pvm for rewriting and parallelizing the RAMS code as part of my PhD thesis work. We at the University of South Carolina have an Intel Paragon A4 and eagerly await the pvm port for that machine so that we can initially port our pvm codes with few if any modifications to our 56 node machine and then to the 512+ node machine at Oak Ridge. Other people here have been using pvm on other atmospheric and ground water models as well and also await the port. How many and what kind of machines have you used with PVM? We mainly use a subnet of 7 IBM RS/6000 workstations but are getting a new bunch of SGI equipment that may be used, mostly for image processing. Also, as I have said, we want to use it extensively on the 56 and 512+ node paragons. Any guess as to when that port will be available? Any comments and suggestions for improvement are welcome. I have a few questions on related topics. I was considering writing a third party PVM tutorial type book. I have had to give introductory type mini courses and I am trying to keep all notes and examples I use for part of this endeavor. What permissions or approvals do I need from your team? I may also need some assistance in proofing and in critiquing and was wondering if we could work something out for this as well? We also are still heavily using pvm 2.4 mainly because of the name changes in 3.x versions. Do you suspect any more such renamings in later upgrades? Also, how long will the pvm software be supported as it is now? I think pvm is a great package and thanks in advance for you help with all my questions. Al Bessey bessey@milo.math.scarolina.edu (803) 777 - 9333 ------------------------------------------------------------------------------- From: arita@mrit.mei.co.jp >Are you using PVM today? Yes. >What are your application(s)? Ising Model. >How many and what kind of machines have you used with PVM? Next , SparcStation2 , SparcStation10 , SolbourneSeries5 arita@mrit.mei.co.jp Panasonic/Japan ------------------------------------------------------------------------------- From: frank@photon.ansto.gov.au (Frank Crawford) > > Are you using PVM today? Yes and no. I've installed it tested it (PVM 3.1) and am waiting on a colleague to start a project using it. > What are your application(s)? The planned applications are all demonstration projects. Primarily comparison of vectorised performance with parallel performance (on both a CM-5 and under PVM). The initial project will be for a diffusion code. > How many and what kind of machines have you used with PVM? Primarily on SGI's and Sun's, however I have previously ported PVM 2.4 to Fujitsu VP2200, i486's running SVR4 and a 68020 based SVR3 system. (I'll port PVM 3.1 to these in the near future, when the colleague mentioned above finds some time to use them.) > Any comments and suggestions for improvement are welcome. Making porting easier. Particularly, the entries in `archcode.c'. While the entries themselves are easy, patching the file for each change on each host gets to be a pain. Some method of updating the types that doesn't need to be compiled in would be nice. > Your help is much appreciated. Your welcome. Frank Crawford ------------------------------------------------------------------------------- From: "Dano Kral" > > Are you using PVM today? We are not using PVM now. > > What are your application(s)? There are calculations in quantum chemistry. > > How many and what kind of machines have you used with PVM? We have only convex C3400 now, but on end of year we will have two HP 735 procesors and we wont create meta cluster C3400 + 2 HP. Regards, Dano. ------------------------------------------------------------------------------- From: j011@cray.op.dlr.de (Leisen Siegfried) 1. I'm not using PVM today in production 2. This is a German research lab and the people are thinking about distributed computing environments in general. They have a few workstations in the computing centre and a lot of Suns and RS6000s in the different institutes. 3. Our first test of PVM was running a mandelbrot set on a CRAY Y-MP and a RS6000/580 and a RS6000/550. I'm a software site analyst from CRAY Research and the people at DLR are playing around with PVM. I did only the installation on the CRAY system. Regards Siegfried Leisen Siegfried Leisen, CRAY Research GmbH e-mail: j011@sn1414.op.dlr.de c/o: DLR Rechenzentrum lei@crmunich0.cray.com Muenchner Str. 20 j011@vm.op.dlr.de 82234 Wessling phone: +49 8153 2032 Germany ------------------------------------------------------------------------------- From: worsch@ira.uka.de pvm> Are you using PVM today? No. pvm> What are your application(s)? We used pvm for moving data from a process running from a machine with dedicated hardware to a process running on another machine. pvm> How many and what kind of machines have you used with PVM? We used pvm only on Sun4 machines. pvm> Any comments and suggestions for improvement are welcome. We would have welcome better documentation (e.g. on which library routine uses which data transportation mechanism). pvm> Your help is much appreciated. You're welcome. Have a nice day. Thomas Worsch ------------------------------------------------------------------------------- From: awidmann@ifp.mat.ethz.ch > Are you using PVM today? Not today but I was using it several weeks to become familiar with it. Unfortunately we still don't have NFS installed why it is a little bit troublesome to work with PVM. As soon as we will have installed the network services NIS and NFS, I will do productive work with PVM. > What are your application(s)? Prediction of properties of polymers by computer simulation. Some of the algorithms we use (developed at our institute) are very suitable for parallelization (e. g. on Alliant or Sequent Symmetry systems). > How many and what kind of machines have you used with PVM? I have been using PVM on our 9 SiliconGraphics workstations (Crimsons and Indigos). As soon as NFS works fine, I will also try to use it on IBM AIX/6000 systems. Regards ------------------------------------------------------------------------- Albert H. Widmann Institut fuer Polymere "The best way ETH Zentrum, CNB E98.2 to predict the future CH - 8092 Zuerich is to invent it." Tel: +41 1 256 5672 FAX: +41 1 261 0215 Alan Kay e-mail: awidmann@IfP.mat.ethz.ch ------------------------------------------------------------------------- ------------------------------------------------------------------------------- From: yaniv@bilbo.fiz.huji.ac.il (Yaniv Romem) I am not using PVM on a regular basis, I have used it in the past few months a few times in order to construct and test parallelizations of programs which were afterwards run on machines like the CM-5. The kinds of applications are generally physics programs. I have used configurations of DEC's (MIPS) , HP's (720) and SUNS (various) though I usually ran on 1-3 computers at a time. By the way I am working on an implementation of PVM on a local parallel computer we have here at the Hebrew University (called the MAKBILAN). I am practically done, although I had to cut out some of the more exotic functions like signals because of the architecture of the machine. Best wishes, Yaniv Romem ------------------------------------------------------------------------------- From: lapique@siisg1.epfl.ch (Lapique) > > Are you using PVM today? Yes > > What are your application(s)? > LU Decomposition, matrix product > How many and what kind of machines have you used with PVM? > 20 : Sun /FDDI, Silicon Graphics/Ethernet Cray-2 > Any comments and suggestions for improvement are welcome. > improve FDDI integration. -- Francis Lapique-EPFL/Computer Center ----------------------------------------------------------------- Swiss Federal Institute of Technology-Lausanne E-Mail : lapique@sic.epfl.ch/lapique@siisg1.epfl.ch Voice-Mail : (41)-21-693-45-96 Fax: (41)-21-693-22-20 ----------------------------------------------------------------- ------------------------------------------------------------------------------- From: fujita@hip.atr.co.jp >> Are you using PVM today? No. >> What are your application(s)? Only example, now. But maybe speech recognition. >> How many and what kind of machines have you used with PVM? 10 Sun's. >> Any comments and suggestions for improvement are welcome. I am studying about PVM and parallel computing. - Fujita ATR Human Information Processing Research Labs. ------------------------------------------------------------------------------- From: fadb@corona.oslo.dnmi.no (Dag Bjoerge) Yes, I'm using PVM. Until now I have only run some small tests. We are going to implement the Nordic+ weather forecasting system HIRLAM in PVM. I have just installed PVM on my SGI, and done some testing. I have done some programming on an Intel Paragon using their 'nx' programming system, but PVM on the Paragon has given me some problems that I have not yet solved. I am a to novice PVM user to give you any valueable suggestions, but I send you my best wishes for your project. Dag Bjoerge The Norwegian Meteorological Institute fadb@corona.oslo.dnmi.no ------------------------------------------------------------------------------- From: "Craig A. Lee" > >Are you using PVM today? Yes. >What are your application(s)? (1) Parallel discrete event simulation (2) Cloud depiction and forecasting, (via CC++). >How many and what kind of machines have you used with PVM? LAN of suns and SGIs (iris), CM-5, KSR, and the iPSC/860 and Touchstone Delta (when I get around to finishing the config of g++) >Any comments and suggestions for improvement are welcome. >Please send the information to pvm-survey@cs.utk.edu. >Your help is much appreciated. I eventually will need to have asynchronous receive (ala the Intel hrecv) for some algorithms. I know that mcavende@ringer.cs.utsa.edu (Mark Cavender) has worked on this for some machines/OSs. Your welcome, --Craig ------------------------------------------------------------------------------- From: Peter Junglas => Are you using PVM today? => yes, I'm just preparing a talk on the Convex Cluster Software (for Convex and HP machines), which includes PVM. => What are your application(s)? => Since I'm working at a computer centre, I have no applications myself, but use PVM for teaching. People here at the TU use PVM for several engeneering applications such as finite element methods, simulation of semiconductors and image processing. I will forward your letter to them. => How many and what kind of machines have you used with PVM? => At the computer centre PVM is installed on a Convex C3840, a Convex C120, a Silicon Pool (3x Indigo and 8x 4d) and a H