The National Academy of Sciences (NAS) interim report on the future of supercomputing is one of three government reports expected to influence federally-funded research in high-end computing.

Taken together, the three reports are expected to reinforce current work in clusters of systems based on off-the-shelf processors while bolstering R&D in custom technologies that could power petaflops-class systems. In another outcome of the studies, government researchers are expanding their plans to acquire high-end systems capable of performing 100 teraflops.

"The message is there is no single solution. There are a diversity of applications," said Susan L. Graham, computer science professor at the University of California, Berkeley and co-chair of the NAS Future of Supercomputing group that issued its interim report Tuesday.

The report, which is not expected to be complete until late next year, warned government budget makers against swinging away from a current focus on off-the-shelf based systems back to custom-built supercomputers.

"Balance is needed between exploiting cost-effective advances in widely used hardware and software and developing custom solutions that meet the most demanding needs.... Continuity and stability in the government funding of supercomputing appear to be essential to the well-being of supercomputing in the United States," said the report.

Graham said researchers should not overreact to NEC Corp.'s Earth Simulator that blindsided many in the high-performance computing community eighteen months ago by delivering a custom-built system five to seven times more powerful than the more off-the-shelf clusters developed in the U.S.

"The Earth Simulator was a platform well suited for some climatic problems for which it was designed," Graham said.

"Both clusters and custom machines will continue to play a role. New programming models and algorithms don't come along every two or three years," added Marc Snir, a professor at the University of Illinois at Champaign-Urbana and the other NAS study co-chair.

Nevertheless, a separate interagency report to be completed for the Bush administration in August is expected to include plans for a fresh source of funding for R&D in petaflops-scale computing and custom architectures. The High-End Computing Revitalization Task Force (HECRTF), will provide a five-year plan for funding supercomputing research beginning in fiscal year 2005.

Separately, a group of high-level government advisors, the so-called Jasons, will issue a third report in August that in effect reviews the government's Accelerated Strategic Computing Initiative (ASCI) program. That program has delivered a range of the world's most powerful systems based on clusters of off-the-shelf microprocessors being used at various national labs to simulate nuclear weapon tests.

The Jasons report, now in a final draft stage, generally reaffirms the ASCI program's targets of acquiring a 100 teraflop system by early 2005 and a petaflop system within ten years, said Mark Seager, an assistant department head for advanced technologies at Lawrence Livermore Laboratories.

While the Jasons did not address architectural issues, they did recommend ASCI planners begin to shift from spending only on high-end, single user so-called capability machines to begin budgeting for capacity machines—high-end systems that could run multiple applications for multiple researchers.

"They said the crunch in capacity systems is significant," said Seager.

ASCI planners agree and are already drafting plans to split their $75 million to $100 million annual systems budget equally between the two system types by 2008. That means the government will be acquiring a broader class of 100-teraflops computers than previous thought, Seager said.

Separately, Seager said he welcomed the prospect of increased government funding of petascale computing through the HECRTF plan. "At some point you have to come up with a new architecture," he said.

Seager said the ASCI program needs a petaflops system by about 2010. Currently, it's not clear whether his work with IBM on the BlueGene/Light program or the Defense Advanced Research Projects Agency's work toward a petaflops system would be able to meet that goal, he added.

Seager praised the recognition in all the reports of the importance of supercomputer R&D in the U.S. "The tough part is to figure out how to make a limited investment and have a big impact on a fairly large computer industry," he said.

By: DocMemory
Copyright © 2023 CST, Inc. All Rights Reserved