DOE seeks super CPU funding

The Clinton administration is considering an Energy Department proposal to invest up to 100 times more money over the next five years in high-performance computers to process complex computer models.

DOE Undersecretary Ernest Moniz outlined the Scientific Simulation Initiative (SSI) last week at the SC98 supercomputing conference in Orlando, Fla. The program would fund the purchase of teraflops computers, which would be capable of proc-essing more than 1 trillion floating-point operations per second, as well as the development of more powerful networks to link them together and the creation of software applications to run on them. The systems initially would be used for advanced climate modeling and designing more efficient combustion engines.

"Clearly we view this as an important opportunity for building on the [Accelerated Strategic Computing Initiative] foundations,'' Moniz said.

The ASCI program has funded development of the first teraflops systems for simulating nuclear weapons tests, and some observers have called the SSI project "Civilian ASCI." One goal of SSI would be to deploy nationwide networks capable of transmitting 1 terabit/sec to allow scientists to share the data processed by the high-speed computers.

"DOE is, at its core, a science and technology agency," Moniz said, and part of the department's mission involves addressing "environmental challenges." For example, he said, investment in more advanced systems would help the United States meet its obligations under an international treaty on global climate change that aims to reduce greenhouse gas emissions. The administration last week signed the United Nations accord on global warming that was negotiated in Buenos Aires, Argentina, where the Clinton administration has been pushing for developing countries to share more of the pollution-reduction burden with industrial nations.

"We can't reliably today predict the regional consequences of global change" without more robust computer models, Moniz said.

Last spring, President Clinton said he planned to heed the advice of the President's Information Technology Advisory Committee (PITAC) and request more money for computer research. The panel, however, recommended putting an extra $1 billion toward long-term basic research, rather than short-term applied research that would be the main thrust of SSI.

"The only thing that [SSI] lacks, in my perspective, is it does not have enough of the fundamental research that PITAC recommended," said David Cooper, a panel member who is chief information officer at DOE's Lawrence Livermore National Laboratory. He added that the program was "not intended to answer PITAC" and was "put together before we made our recommendations."

A DOE spokeswoman said the department is negotiating with the Office of Management and Budget about a budget for the program, which is likely to involve other federal agencies besides DOE. Cooper said DOE and its laboratories have been working with the National Science Foundation "to put together a program that jointly utilizes our various strengths."

A draft overview of the project posted Nov. 10 on the DOE World Wide Web site outlines four funding scenarios, ranging from no new funding to $600 million over the next five years for hardware acquisition alone, but the draft does not recommend any of the funding scenarios. Supplying a 1 terabit/sec network could cost "as much as $80 million per year,'' according to the document.

Some of the money could go toward grants for universities. University researchers, led by the NSF-backed National Computational Science Alliance, have been pushing for funds to build a superfast national academic research network. "This is a tremendous opportunity,'' said John Toole, deputy director for alliance programs at the National Center for Supercomputing Applications, Champaign, Ill., which heads the group.

Toole said he is "bullish" that the administration will propose more computer research funding next year, but the balance that will be struck between basic and applied research is not clear. "It's important for the administration to put together a clear, strong view of how these things fit together,'' he said.