Oak Ridge National Laboratory
Measuring Up
On the Petascale
“Anyone can build a fast
CPU. The trick is to build a fast
system.”
— Seymour Cray
The words of Seymour Cray, founder of supercomputer maker Cray Inc., will soon be tested. The company is to deliver a system to Oak Ridge National Laboratory (ORNL).
Petascale computing has the potential to eclipse today’s fastest computer by a factor of three. But how fast it really runs depends more on the lab than on the manufacturer.
As the Department of Energy’s National Leadership Computing Facility (NLCF), Oak Ridge has been asked to take a dramatic step toward a new capability for complex, computationally intense science.
“Leadership Class Computing is a combination of world-leading computing power and the policies that allow it to be used to solve very large problems,” says Doug Kothe, director of science at the lab’s National Center for Computational Sciences (NCCS). Instead of hundreds or thousands of simulations running on the system at any one time, there may be only a handful. The idea is to push the limit on what simulation can tell us about about the natural world.
Jaguar, Oak Ridge Lab’s Cray XT4 computer system Click image for larger version and more information |
The new Cray “Baker” system will make that look like small potatoes. When it arrives in early 2009, it is expected to achieve petascale speed — more than a quadrillion operations per second, 10 times Jaguar’s current Linpack rating.
But fast computers alone can’t guarantee science breakthroughs. It’s just as important to allocate large blocks of time on them to attack the most difficult problems. DOE does that through the Innovative and Novel Computational Impact on Theory and Experiment (INCITE) program, which provides large computer time grants to just a handful of peer-reviwed projects. The 2007 program made 28 allocations, but only a handful will be selected to have the first crack at Baker.
“It’s an interesting and exciting new model, where you are trying to go deep into the science, and that can be done if you only have a few projects,” Kothe says.
A third ingredient, besides fast computers and large allocations, also is necessary if these projects — some of the most complex calculations ever attempted — are to succeed. The NCCS works with the science teams through user assistance and scientific computing groups. A 5-million processor-hour INCITE allocation is equivalent to a multi-million-dollar grant in terms of the support they receive, Kothe explains.
“To do breakthrough science at a user facility, we can’t just be a ‘cycle shop’ and say ‘OK, here’s your account, good luck,’ ” Kothe says. “We intimately involve ourselves with each team. We partner with them in ways that can help them by giving them our best and brightest people who are most aligned with their interests and needs.”
That includes former DOE Computational Science Graduate Fellowship scholar Richard Mills. Mills worked in user support at Oak Ridge after completing his practicum with Peter Lichtner at Los Alamos National Laboratory. That experience led to a full-fledged collaboration with Los Alamos, and now Mills is a fully supported member of ORNL’s computational earth sciences group.
“Richard spun up very quickly through our system,” Kothe says. “It’s a good example of what’s possible for computer scientists who join our group.”
