The Great U.S. Supercomputer

Competition with fancy machines and wads of cash, state schools steal national center scientists Christmas came early in 1985 for serious number crunchers. In the spring of that year, the National Science Foundation christened five national supercomputing centers and sent them forth into the world to meet the grand challenges of science and engineering. The NSF's idea was to fund these silicon meccas so that they could maintain state-of-the-art technical facilities and provide supercomputing services to a nation of users. With top machinery and personnel, these five centers, located at the University of Pittsburgh, the University of Illinois, Princeton, Cornell, and the University of California, San Diego, were to stimulate world-class research and analysis in everything from pure mathematics to fluid dynamics, helping to bolster U.S. leadership in science and technology. With federal money behind them, the centers soon had the biggest, fastest, and fanciest machines avail- able. This primacy, however, turned out to be demanding, expensive, and, some fear, ephemeral. Today, all five centers—only three years after their creation—anticipate serious difficulties if present trends continue. Most of them have yet to receive the full funding promised in the original five-year cooperative agreement with the NSF. In March, the centers’ directors testified before a Congressional subcommittee that budget cuts were seriously undermining their ambitions and programs. One director, Sidney Karin of San Diego, warned in his testimony: “I must report to you that we are having difficulty staying up at the leading edge.” Karin isn’t just crying wolf. At the current level of funding, the national centers are finding it difficult to expand, upgrade, and provide adequate ancillary support to meet the needs of their supercomputing user base. And that has created a second threat to their primacy: competition for supercomputing superstar status from brash state universitites wielding some mighty fancy hardware and flashing thick wads of cash. Just last March, Ohio State, in an ambitious raid, snatched Alison Brown away from the Cornell Theory Center where she was associate director for networking and advanced computing. With her came Nobel laureate Kenneth Wilson, director of the Theory Center and Brown’s husband. The principal lure for Brown: Ohio State offered a supercomputing center with strong financial support and a clear future. The facility intends to spend $22 million that the Ohio legislature appropriated in April to add Cray Research’s newest model—the Cray Y-MP—to its recently acquired Cray X-MP The Y-MP is to be installed next year; no NSF center currently has one. Charles Bender, director of the Ohio State facility, explains his successful raid on Cornell bluntly, “When your toys aren’t the best in town, you start to lose your people.” And Ohio State is not alone in staking a claim to supercomputing king-of-the-hill status. When Purdue added an ETA 10, manufactured by Control Data’s ETA Systems, to its Control Data Cyber 205 last month—nearly doubling the school’s supercomputing power—it became one of a very few schools with more than one of the big machines on campus. And the University of Minnesota—at 10 years old the grandmother of all the supercomputer facilities—has three supercomputers. One of them, a Cray-2 that the school is in the process of upgrading, has more memory than all the national centers’ supercomputers combined. We feel we are already the leading academic supercomputing center in the country,” says Minnesota acting director Donald Thihlar. Meanwhile, the University of Kentucky’s fledgling facility has snapped up the former director of NSF’s supercomputing program, John Connolly, who was hindered by federal conflict-of-interest regulations from taking a position at one of the NSF centers. While not every state center is ready to take on the NSF centers, the rivalries between the two groups are heating up. “I think there is no doubt that is the case,” William McCurdy, former director of the Ohio State center, says. “We’re competing for people and resources.” While the state facilities are clearly in the ascent, few observers are willing at this point to count out the NSF-anointed centers. Nevertheless, the tale of their decline can be taken as a lesson in the perils of hitching an expensive technology to the government’s funding horse. The basic problem with supercomputers is that, with all their power and speed and memory, they have the lifespan of a lab rat. After two or three years, the supercomputers at the national centers are middle-aged. And here, as in many instances, the federal funding horse turns out to be more nag than thoroughbred. To regain their solid position at the forefront of supercomputing, it is clear that the NSF-backed national centers must replace their machines with the next generation of mammoth number crunchers within the next few years. But this will be difficult, directors of the five centers say, unless they receive their full budget request for 1989—$54 million—and unless the NSF adopts a multiyear funding scheme that allows for upgrade planning. Since 1985, the national centers have failed to receive the full amount of funding promised them in the original multiyear cooperative agreement. Cornell, for example, has received only 69% of the money it originally expected. To make matters even more difficult, NSF dollars—besides being in short supply—also carry with them some stifling restrictions They cannot be used to create incentives for interdisciplinary collaborations, Cornell director Wilson says. He was thwarted in his effort to encourage people to share numerical expertise across different disciplines. Wilson notes that at Ohio State, no constraints to such scientific creativity exist. Brown and Wilson cite a number of personal and professional reasons for leaving Cornell. Brown was disappointed last fall when the NSF-supported Cornell center—in collaboration with private telecommunications developers GTE and Sprint—lost a networking contract with the NSF itself to a consortium composed of IBM, MCI, and eight state-supported universities in Michigan. That’s when Brown began casting about for new challenges, she says. The fledgling Ohio State center caught her eye because it “had been set up in a way that I thought was just what ought to be done, technically,” she says. The facility uses UNICOS, the Cray operating system, which Brown calls “the wave of the future.” Once Brown was theirs, Ohio State set out to make Wilson feel at home, even arranging a meeting with Ohio’s Gov. Richard F. Celeste. Wilson, who will be working on electronic structure with Ohio State’s Isaiah Shavitt and John Wilkins (another Cornell defector), emphasizes that he opted to leave Cornell because of “the spectacular job opportunity for my wife and the ability to participate in a center comparable to the Theory Center at Cornell.” Ohio’s push for national prominence, he says, “will put pressure on the national centers that is not there now. And competition, in my view, is always a good thing.” Yet, it’s too early to count the NSF centers out. While federal support has its downside—as evidenced by the current raft of budget complaints—it also has advantages. When their future looked brighter, for example, the national centers were able to strike mutually beneficial deals with prestigious computer manufacturers. Cornell, for instance, paid nothing for its IBM 3090 mainframe; instead, it agreed in return to work with IBM on expanding the 3090’s capabilities, particularly in the area of parallel processing. Then there’s the advantage of clout. “Being an NSF center gives one enormous clout, which you don’t get when you’re a state center,” says Wilson. His point is supported by the bitter tale of Judson Harper, vice president of research at Colorado State. In 1984, Colorado State, along with Minnesota and Purdue, were chosen as national centers in the first phase of the NSF program; but they were left out in 1985 with the introduction of the second phase. Says Harper. “There’s a lot of things you can do when you have something to leverage off of. We just lost our fulcrum.” The national centers, however, have been able to attract more than Congressional support; to the envy of their state-supported siblings, they have obtained substantial industry support. Illinois, for instance, announced last month that Amoco, Eastman Kodak, Eli Lilly, and Motorola will contribute a major portion of the money for the purchase of a new Cray- 2S, to arrive this fall. But directors of the national centers fear that if one of their major partners—the federal government—reduces its support, industry may shy away. There is one group that will be hurt if the national centers decline, no matter how good the state centers become. Those are the scientists who live in states without supercomputing centers and whose only access to big number-crunching is through the national centers. “If you’re from a tiny, have-not state like Massachusetts then you rely on connecting onto other people’s supercomputers,” says Connolly. And that means the national centers because, in general, the states won’t permit out-of-state time-sharing. It is this special role for NSF centers that leads some to wonder whether science in this country doesn’t need both state and national centers. Ohio State’s Brown envisions some of the state centers becoming “centers of excellence,” distinguishing themselves with scientific innovation, while the NSF centers continue to serve the computing needs of a large, national audience. And former NSF director Connolly is careful to sound conciliatory: “It’s not us versus them; it’s all of a piece.” Just how the pie will be sliced may be decided with the fiscal-year 1989 budget allocations. In their March testimony before the House Subcommittee on Science, Research and Technology, the directors called for full funding of the original $54 million request. The subcommittee responded by authorizing a surprising $74 million, with the additional $20 million, intended for hardware upgrades, taken from an NSF request for $150 million for science and technology centers. As of mid-May, the House Committee on Science, Space and Technology had authorized $74 million as a floor for the supercomputer centers. Meanwhile, the House Subcommittee on Housing and Urban Development and Independent Agencies appropriated $54 million for the centers. Melvyn Ciment, acting director of the NSF program, would be happy even without the $20 million in extra funds. Says Ciment: “I’m interested in getting the money that was requested. That’s the critical step.” But if the request is not met, then the very existence of the national centers is called into question. Without detailing the possibilities, Ciment hints. “If you don’t have the money for five, you have to start thinking of other numbers.” “It’s not easy to throw one of your children overboard,” he adds. “We’ll hold out as long as we can.”

The Great U.S. Supercomputer

Interested in reading more?

Become a Member of