NSF Supercomputer Center Sites are Announced

By Louise Arnheim

First there were five, now there are four, and soon there will be two. In late March, the National Science Foundation (NSF) announced it would discontinue funding two of its four supercomputer centers: the Cornell Theory Center and the Pittsburgh Supercomputing Center (PSC).

The two remaining centers-the National Center for Supercomputing Applications (NCSA) at the University of Illinois at Urbana-Champaign (UIUC) and the San Diego Supercomputer Center at the University of California at San Diego-will continue receiving federal support through a new program, Partnerships for Advanced Computational Infrastructure. PACI is billed by NSF as the next phase of high-end computing.

Following a three-year process of program review and proposal submission, NCSA was awarded funding for its National Computational Science Alliance (the Alliance) partnership, and the San Diego Supercomputer Center got the go-ahead for its National Partnership for Advanced Computational Infrastructure (NPACI).

The Alliance is led by Larry Smarr, current director of the supercomputer center at UIUC, and includes the following institutions as major (proposed) partners: the California Institute of Technology, the University of California at Berkeley, the University of California at Los Angeles, the University of Maryland, the University of Michigan and the University of Texas at Austin. Together these institutions will prototype a "National Technology Grid."

According to a press release, the grid "will serve as an early model for a full-scale Advanced Computational Infrastructure, which will be built by the computer, communications and software vendors to support our nation's computational scientists and engineers in academia, industry and government." The Alliance is organized around four sets of teams:

• The Application Technologies Teams, which will focus on the following six science and engineering areas: cosmology, chemical engineering, environmental hydrology, molecular biology, nanomaterials and scientific instrumentation. Funding for the basic scientific research will come from team members, an NCSA press release said.

• The Enabling Technologies Teams, which will study tools for parallel computing, heterogeneous computing, and data and collaborative computing.

• Regional Partners, which includes the Committee on Institutional Cooperation, the Southeastern Universities Research Association and the Experimental Program to Stimulate Competitive Research.

• Education, Outreach and Training Teams that will help users "underserved by the commercial sector"-such as K-12 students, minority groups and government agencies-become more familiar with advanced computational technologies.

NPACI, which is headed up by Sidney Karin, the founding director of the San Diego Supercomputer Center, will focus on three areas: providing a metacomputing environment to "link high-performance, geographically separated computers of different architectures over very-high-speed networks," facilitating data-intensive computing and bringing "nontraditional" groups (minorities and women) into high-performance computing.

NPACI will concentrate outreach efforts in California and Texas, which, according to NSF, account for 20% of the country's elementary and secondary students.

Included among this "partnership of partnerships," as NPACI calls it, are: NSF, Energy Department and NASA Grand and National Challenge projects; NSF, Defense Advanced Research Projects Agency and NASA Digital Library projects; National Institutes of Health Research Resources; Energy's Lawrence Berkeley, Lawrence Livermore, Los Alamos and Pacific Northwest national laboratories; and NASA's Jet Propulsion Laboratory. NPACI's partner sites include the California Institute of Technology, the University of Texas, the University of Michigan, the University of Maryland, the University of California, the University of Virginia and Stanford University.

PACI funding will be about $65 million annually, which, according to NSF, is "similar to previous levels." Although PACI is designed for a 10-year run, the cooperative agreements for the two partnerships are for five years. An independent review of the program will take place during the fourth year of operation.

At a recent press conference announcing the PACI partnerships, Karin said the new program would enable scientists to "seamlessly acquire data from instruments from one place, massage it at another and analyze it at yet another." He said the new effort was a "much more integrated, much more broadly based-program."

Also at that press conference, Smarr suggested using the term "superinfrastructure" instead of "supercomputer" to talk about the future. "That's what the next 10 years are about," he said.

The Pittsburgh and Cornell supercomputing centers will be given a year's operating budget and two years to phase out operations. However, both the University of Pittsburgh (along with Carnegie Mellon University) and Cornell University have announced plans to keep their respective centers operating and to seek funding from alternative sources.

The actual mechanics of the transition and how it will affect current users, however, are unclear. Current users of the Pittsburgh facility include researchers at the Center for Analysis and Prediction of Storms (based at the University of Oklahoma), scientists at the Los Alamos National Laboratory who use PSC for modeling of the Earth's magnetic field, pharmaceutical chemists at the University of California at San Francisco working on DNA simulation and an astrophysics group at the University of Arizona conducting modeling of supernovas.

The Cornell Theory Center serves 2,000 users. According to a center spokesperson, anywhere from half to two-thirds of those users will be affected by the transition.

What made the Alliance and NPACI proposals "winners?" According to Robert Borchers, NSF division director for Advanced Scientific Computing, "in essence, they 'got it.'" Both proposals, he said, encompassed a vision of distributed computing and partnering, and both "are highly complementary." They went "beyond the vision," he added.

"The quality of the proposals we received from NCSA and NPACI represent a breadth of vision beyond what we had even hoped for," said Paul Young, senior adviser to NSF's Computer and Information Science and Engineering Directorate. "The proposals expanded the roles and impact of the leading-edge sites. The partnerships will maintain the country's lead in computational science. They will further the use of computers in all disciplines of research and offer new educational opportunities for people ranging from kindergartners through Ph.D.s."

Computing Research News learned that during the proposal process, NSF had suggested to PSC that it partner with UIUC. However, because the arrangement would have required the Pittsburgh facility to move its computers (as well as several staff members) to Illinois, PSC did not pursue that option. In addressing a reporter's question about closing the Pittsburgh supercomputer center in particular, Karin said the center "has been a valuable resource" and noted that the PSC had formed "a strong bond" with Cray Research Inc.

Similarly, Smarr noted that both Pittsburgh and Cornell had "provided real service to the country" over the past 12 years, but that the issue was "simple arithmetic." The fact that federal funding to at least one supercomputing center would be cut apparently has been anticipated for some time. Looking forward to the new program, Smarr said, "Supercomputers are not being de-emphasized, but re-emphasized."

Smarr also noted how the closing of the John von Neumann Supercomputing Center (at Princeton University) eight years ago occurred in a shorter time period than that allotted for PSC and the Cornell center.

The recent announcements are the culmination of an evaluation process that began three years ago. In fall 1994, the National Science Board, which oversees NSF, instructed the foundation to review the supercomputer centers. A few months later, a task force chaired by Ed Hayes of Ohio State University was convened to study these issues. The NSB used the Report of the Task Force on the Future of the NSF Supercomputer Centers Program to develop future plans.

During a pre-proposal round in spring 1996, 10 applicant bids were reviewed. These pre-proposals were reviewed and returned to bidders a month later. In September 1996, NSF received six PACI proposals (four from the existing supercomputer centers; two from other institutions).

A panel of experts narrowed the bids down to the four supercomputing centers, and site visits were set up for October. Then, beginning late last year, a panel met to make recommendations to NSF, which in turn made its recommendation to NSB. On March 28, the two winning partnerships were announced. NSF's Supercomputer Program began in 1985; five centers were established at that time. Funding for the current program ends this fiscal year, with PACI program funding beginning in fiscal 1998.