NSF & Congress
Science Committee Holds Hearing on U.S. Leadership in Supercomputing
July 16, 2003
The House Committee on Science held a hearing July 16 to examine whether the United States is losing ground to foreign competitors in the production and use of supercomputers and whether federal agencies proposed paths for advancing U.S. supercomputing capabilities are adequate to maintain the U.S. lead.
In particular, committee members expressed concern with the first-place position of the Japanese Earth Simulator on the list of the world's Top 500 supercomputers, whether federal agencies were coordinating their activities, and specifically about the National Science Foundation's plans for high-end computing programs.
Chairman Sherwood Boehlert (R-NY), in his opening remarks, noted that the U.S. is not "at a point of crisis," but is at a pivotal point in maintaining U.S. leadership. He said he fears that "lethargy is setting in" and that much is at stake. Ranking minority member Ralph Hall (D-TX) reiterated the hearing's emphasis on assessing how agencies were providing high-end computing (HEC) resources and whether commodity component systems were the best path to compete with the Earth Simulator.
Dr. Raymond L. Orbach, director of the Department of Energy's Office of Science, testified that
the Bush administration has forged an interagency plan for high-end computing. DOE's Office of Science is cooperating with the National Nuclear Security Agency on the development path for a next-generation supercomputer and has a memorandum of understanding under review with Defense Advanced Research Projects Agency. Dr. Orbach stated that the Office of Science expects to play a major role in implementing the recommendations of the Office of Science and Technology Policy-convened High-End Computing Revitalization Task Force (HECRTF).
Dr. Peter Freeman, assistant director for Computer and Information Science and Engineering at NSF, emphasized in his testimony the concept of cyberinfrastructure, which signifies a set of integrated high-end facilities including computers, networks, storage and instruments. According to the recent report from NSF's Advisory Committee on Cyberinfrastructure, cyberinfrastructure promises to revolutionize research, and NSF, in partnership with other agencies, is creating and deploying an advanced cyberinfrastructure to continue U.S. leadership.
Dr. Freeman stated that NSF is committed to expanding its support for high-end computing and following through on recommendations from the cyberinfrastructure report. NSF will announce its plans in the coming months. In addition, NSF is committed to implementing the recommendations of the HECRTF.
Dr. Daniel A. Reed, director of the National Center for Supercomputing Applications, testified that
NSF has played and must continue to play a critical role in providing community access to HEC. The planned cyberinfrastructure will accelerate scientific discovery. At issue, though, was the level of investment. He noted that grids are not a substitute for high-end computers. Dr. Reed stated that the U.S. needs long-term, sustained investment in software as well as hardware, and that new HEC designs will be needed to avoid relying on hardware too narrowly focused on commercial needs.
Mr. Vincent Scarafino, manager of numerically intensive computing at Ford Motor Company, described Ford's uses of HEC to accelerate the design cycle of its products. Mr. Scarafino testified that through the mid-1990s, the federal government's support of new computer architecture development for science and security helped to advance the state of industry. Starting in the mid-1990s, however, the government has followed the path of parallel processing. The hardest problems do not adapt well to parallel processing and industry has not made many advances in new applications. He stated that the government cannot
rely on industry to drive fundamental computer design and that "U.S. leadership is at risk." While the needs of science and industry are not diverging, both have been forced to adapt non-optimal hardware to their needs.
Chairman Boehlert's questions first focused on issues of coordination both through the HECRTF and the National Coordination Office for Information Technology Research and Development (NITRD). Both Drs. Orbach and Freeman emphasized support for the various coordinating mechanisms in place. Mr. Boehlert questioned Dr. Freeman about funding for HEC beyond FY 04, and Freeman noted that overall funding for HEC would be increasing and that Mr. Boehlert should be optimistic about the funding for the supercomputer centers beyond FY 04. Boehlert asked Dr. Reed whether the U.S. was moving in the right direction, to which Reed reiterated the need for a long-term research pipeline to design, prototype and demonstrate novel high-end computing architectures.
Rep. Randy Neugebarger (R-TX) asked Dr. Reed whether NSF was providing adequate support for the supercomputer centers and if he had seen any change in policy. Reed noted that NSF is a long-time supporter of supercomputing and that budgets are tight, but that "we do see untapped opportunities." Mr. Neugebarger asked whether there was unmet demand for HEC resources, and Reed reiterated that NSF centers have computing power to solve only one scientific grand challenge with existing capacity, so priorities have to be set. Mr. Scarafino noted that Ford has been at a standstill since the mid-1990s as far as capability; a 10-year-old Cray T90 still provides computing capability better than any other system
for some applications.
Rep. Phil Gingrey (R-GA) also asked about funding for the NSF supercomputer centers and NSF's focus on grid computing. Dr. Freeman described current funding for the Partnerships for Advanced Computational Infrastructure (PACI) program and noted that NSF is restructuring how that support is provided, not reducing the amount. In fact, NSF is trying to increase that amount. Mr. Gingrey also asked which agency supports research for
new architectures and software, and Freeman noted that NSF provides about $400 million dollars for fundamental computer science research, about half the nation's investment.