Meeting the Demand for HPC Resources

BY
Sid Karin, NPACI Director

he National Science Foundation (NSF) will celebrate its 50th anniversary in the year 2000. NSF, among all federal agencies, has led the nation in providing high-performance computational and networking facilities to the academic science, engineering, and mathematics communities. The NSF supercomputer centers have been an invaluable part of the country's "digital fabric" since 1985, and the year 2000 will also mark the third year of the Partnerships for Advanced Computational Infrastructure (PACI). The PACI program has already contributed important scientific advances closely tied to the construction of the national computational grid, including pioneering work in data-intensive computing and other enabling technologies.

Today, a confluence of scientific, technical, and governmental circumstances is making possible a great leap forward in computational power and its application to problems of social importance.

A NECESSARY LEAP

Why is such a leap possible? And why is it necessary?

Simulation and modeling have led to revolutionary insights over the past decade. The power of these methods is so great that the defense establishment is preparing to rely heavily on simulation studies as a means of maintaining the U.S. nuclear arsenal.

These studies in turn will take place in the Department of Energy's (DOE) Accelerated Strategic Computing Initiative (ASCI) program, which is providing computing machinery of startling capacity: thousands of processors operating in a massively parallel environment. These will soon produce sustained operations in the 30 to 100 teraflops range or higher.

What are the wider implications of these advances?

In its recently completed report, the President's Information Technology Advisory Committee (PITAC) forcefully draws attention to serious deficiencies in the nation's capabilities in information technology, particularly in the areas of software development, scalable information infrastructure, and high-end computing. As the PITAC report emphasized (see my editorial in the July–September 1998 issue of enVision), these areas are vital to maintaining and increasing our leadership in simulation and modeling.

A new initiative is required to let the PACI program keep up with the demand of the academic science community. A shortfall in computational capacity began about 18 months ago, and it is increasing exponentially. At its first meeting, the National Resource Allocation Committee (NRAC) of PACI was able to allocate only 20% more resources than the old Metacenter Allocation Committee had awarded the year before. Proposals of the highest merit were awarded, on average, only 50% of the resources requested. Many disciplines have emerging scientific challenges that require multi-teraflops capacity, but at the current rate of increase it will be years before NSF will acquire such machines in its PACI centers.

The T in the EOT-PACI program--training--is key to the wider successes of computational science as an engine of productivity nationally. If the PACI facilities fall behind, universities will be in danger of becoming decoupled from work on problems with the scale of parallelism of the ASCI facilities--facilities which must surely rely nonetheless on a stream of people trained in academe.

CLOSING THE GAP

NSF is well aware of the rapidity with which the gap is widening. One proposal, called APEX, suggests a doubling of support to PACI resources. To supply computing capacity at a nearer-to-ASCI level to nondefense researchers, leaders at DOE have conceived a Strategic Simulation Initiative (SSI), with primary thrusts in the areas of climate modeling, combustion modeling, and what DOE terms basic science. It is important to note the large intellectual overlap between these areas and existing NSF programs.

The DOE leadership has approached NSF, NIH, NOAA, and NASA, among others, suggesting a partnership and unified approach to a budget initiative that would begin in fiscal year 2000. The response has been generally positive. The idea is to capitalize on the ASCI program's development of the new supercomputers, in partnership with industry, by building copies for use by the academic and non-defense national lab communities.

With opportunities open, driven by a clear necessity, action to support such initiatives is in order. I will be doing all I can on behalf of a new infusion of resources. I hope you can help in the effort. If so, please get in touch with me at skarin@npaci.edu.