About Me

I am a computational scientist at ORNL. Previously, I was a postdoc working for George Fann in the Computational Mathematics group in the Computer Science and Mathematics Division. I started working here as a postdoc in September 2005, and became a staff member in September 2007.

I obtained my Ph.D. in Computer Science with a certificate in Computational Science and Engineering from the University of Illinois at Urbana-Champaign in 2005. I was a student of Mike Heath in the Scientific Computing group. The title of my dissertation is "The Diffusion Equation Method for Global Optimization and Its Application to Magnetotelluric Geoprospecting." Way back before that, I earned a B.S. in Physics from the University of Kentucky.

I also enjoy teaching and working with students. I originated and still teach the NCCS "Crash Course in Supercomputing" every summer. I also like to hire a couple of student interns each year, so if you're a student interested in working with me (in the summer or during the school year), drop me a line!

These days, I'm working on several different projects:

Scientific Computing Liaison

In my job in the Scientific Computing group, I work with several different projects to get them up and running at the petascale. While the majority of our users have experience running their codes in parallel, until they access our machines, they have never had the opportunity to run on such a large machine and solve such large problems. For my projects, I am available to do anything from tell them the best compiler flags or improve their job scripts, to becoming an embedded member of their code development team.

In addition, I work on the Joule project, in which four codes are selected to be improved in performance over the course of the year, and the results of these efforts reported to OMB. I coordinate the NCCS side of the effort.

Load Balancing at the Petascale

I am currently working with the NUCCOR nuclear physics code team to load balance their code and generally improve its performance. In this case, we have to divide up the work across all the processors in a scalable way. Right now, the code scales up to about 1000 cores, but we need to get it to a point where it can work well on a leadership supercomputer such as Jaguar, ORNL's flagship machine.

I worked on load balancing techniques for the octtrees in MADNESS. On a leadership supercomputer, we want to evenly distribute the load over hundreds or even thousands of processors, while minimizing the number of broken links. Something that makes our load balancing requirements unique is the fact that we utilize the whole tree for some operations, rather than just the leaf nodes. I came up with some ideas that suited our purposes. For more about our project, see our SCIDAC project web page.

Evolution of the Biofuel Supply Chain Infrastructure

I'm involved with this project modeling the biofuel supply chain infrastructure. The premise is, where should we put biorefineries and storage, and what crops should be grown? This boils down to a large mixed-integer linear program. My role is to facilitate the solution of this problem at the petascale.

See my CV (short) (long) for more information about me.

Research Interests

My research interests include

• Inverse and ill-posed problems
• Load balancing at the petascale and above
• Optimization
• Scientific computing
• Supercomputing

Links

• My handy links
• Crash Course in Supercomputing (slides available from 2007 and 2006 courses)
• Numerical Analysis Songs

---

Last updated December 17, 2009

hartmanbakrj@ornl.gov