Oak Ridge National Laboratory

Measuring Up on the Petascale

(Page 3 of 3)

“The Scientific Computing Group is really synergistic, since we are often members of the development teams for the applications and members of the operation team for the center,” White says.  “We act as advocates and proxies within the center for each project, and we bring detailed knowledge of the center computers, infrastructure, and plans back to the development teams.”

White says POP could be among a handful of pioneering applications used to track performance when the Baker system first comes on line.

“The Scientific Computing Group exists because of the understanding that we are treading new ground; this isn’t just using computers as is typically done, but really pushing the envelope on the scale of parallel computing,” White says.

Making the Transition to Petascale

“The NCCS is co-located at ORNL with many of DOE’s premier science teams who use the computer systems,” Kothe says.  Several of these teams are already busy developing codes specifically for petascale machines.  Two such ORNL projects received INCITE awards in 2007.

Robert Harrison, group leader of the Computational Chemical Sciences Group at Oak Ridge, is working with NCCS in an attempt to stave off program obsolescence by working ahead of the game.  He’s developing a quantum chemistry program that incorporates petascale capability into the initial design (see sidebar).

“Many codes have been caught sort of flat-footed in the race to get to the petascale because they have to make a transition from running on a few hundred or thousand processors to running on hundreds of thousands or millions of processors,” Harrison says.  “Typically what happens with older codes is that programmers take an existing program, parallelize it, and get stuck there, limiting their ability to migrate to even larger machines.” Likewise, Thomas Schulthess of ORNL’s Computer Science and Mathematics Division heads a research team that is developing a program called DCA++, a new-generation program that will predict behavior of materials at the nanoscale.  The project’s goal is ultimately to develop new materials designed on the nanoscale.

When these and other researchers are ready to run their simulations, members of the Scientific Computing Group will be there to serve as liaisons by scheduling time on the machine and smoothing the rough edges of the programs.

“Setting up queuing systems to be effective is a black art, particularly when you have really large jobs,” White says.  “Fitting everything in the available space/time holes can be as much a social science as technical.  When a project has a big deadline coming up, we try to make sure they’ve got the resources they need.”

Kothe adds: “We’ve mastered the art of user support, while pushing the envelope of the science that can be done on these machines.  That’s what makes us unique.”

« Previous       1   |   2   |   3   |   References   |   Sidebar   |   Print       Next »