Computing and the National Science Foundation, 1950-2016. William Aspray

Чтение книги онлайн.

СКАЧАТЬ in December 1979, the Advisory Committee recommended that NSF invest $15 million each year in a national competition for resource grants. These grants would “total no less than $250,000 and no more than $2 million,” include maintenance and software support of 10% of the capital costs and be available to individual researchers and departments. The report expected that after five years, the program would “produce at least 25 well-equipped university laboratories among the 64 computer science Ph.D. degree granting universities.”

      John Pasta and Kent Curtis responded to these recommendations by “taxing” the Standard Research Projects Support (SRPS) budget in FY 1980 by $1 million, almost the entirety of the FY 1980 budget increase, to create the Coordinated Experimental Research (CER) program. It had three main thrusts: a CER facilities program, a program to assist the research community in developing networking services in support of computer science research, and grants to attract experimentalists into a university environment. Curtis sent a “dear colleague letter” in November 1979 inviting proposals for what would become the CER facilities program. Program descriptions for a New Investigator program and a Postdoctoral program came after for FY 1981 funding. In the first year the CER program funded one facilities grant and a CSNET study grant.

      Today, electronic “dear colleague letters” quickly gain broad audiences, but in 1979 Curtis mailed a letter to the computer science Ph.D.-granting departments. The timeframe was short and we received only seven proposals. Predictably, most of the proposals came from people familiar with experimental computer science within NSF and the Computer Science Section. I said in a 1990 interview, “in the first set of proposals there was one good proposal, one sort of half-good proposal, and the rest of them were . . . bad proposals [although they involved] some very good people”⁹²—bad in the sense that they seemed to be independent projects “stapled together” rather than a unified coherent proposal.

      Our first challenge was to determine how to review the proposals. We decided on a multi-stage process: mail reviews, site visits, and a final panel. We decided that I with two external reviewers would personally visit the sites of all of the proposing institutions, following receipt of mail reviews. Principal Investigators would have the mail reviews to react to, along with questions by the site visitors. The final panel included no academics but instead the heads of major industry and non-profit laboratories.

      The first CER award was made to the University of Washington to construct the Eden operating system with a goal to build a system coupling the performance of powerful personal machines with the resource sharing and accessing capability of a modern time-sharing system. This major research project involved a majority of the departmental faculty and produced a facility that could support a variety of research projects. The Eden Project attracted co-funding from Intel and Digital Equipment, whose technical staff collaborated with Washington on the research.

      In FY 1981, CER became an official program with $3.6 million dollars of the Special Projects budget identified as “experimental computer science” with other line items for CSNET, young investigator, and postdoctoral awards. As we discuss below, a revised set of CSNET proposals were received and approved by the National Science Board. One postdoctoral award and four new investigator awards were made. The CER program received 24 proposals responsive to the program announcement, which were distributed to other NSF, Office of Naval Research (ONR), and DARPA programs. We hoped for significant DoD involvement in developing CER sites as DARPA, ONR, the Air Force Office of Scientific Research (AFOSR), and the Army Research Office were planning a “Computer Resources Initiative” in FY 1982 with $30 million among the DoD science agencies.

      With cooperation from the DoD agencies, we selected 11 proposals for site visits similar to those of the prior year. From the eleven sites visited, five were discussed with DARPA and ONR. Following a budget negotiation, four proposals went to the National Science Board, which approved three immediately and a fourth later. DARPA eventually funded a version of a fifth proposal. The four new NSF CER awards went to Cornell to support investigation into the programming process, Illinois for the construction of computer aids to program and system development, the University of Wisconsin–Madison for construction of a 50-node network of powerful computing devices, and Yale for facilities to support artificial intelligence and natural language processing, numerical computing, and computer architecture.

      In the succeeding years when I managed the CER program, five awards were made in FY 1982 to Rice, Brown, Utah, UCLA, and Texas. Four awards were made in FY 1983 to North Carolina Chapel Hill, Pennsylvania, Maryland College Park, and Duke. SUNY Stony Brook, Rochester, Arizona, and New York University received grants in FY 1984. After I left for Berkeley, Harry Hedges joined NSF from Michigan State to run the CER program. In FY 1985, Hedges and Bruce Barnes made awards to Princeton, UMass Amherst, Colorado Boulder, and Minnesota.

      As the CER program began, we did not completely agree on its goals. Some supported the concept of “Centers of Excellence”; some supported funding large, multi-investigator “experimental” research projects; and some promoted large-scale facilities grants, which would include equipment, maintenance, supplies, and technical staff. Clearly the Eden Project fell into the large, multi-investigator “experimental” research project category. While I personally favored a focus on large, collaborative research projects, there were few awards in this category. Reviewers prioritized facilities support and grants to institutions with an existing core of potential experimental computer scientists. Almost all of the grants had a unifying theme, but the available funding limited grants to support for equipment, maintenance, and support staff, with some support for the lead principal investigators, and provided few or no funds for graduate students, postdocs, or faculty salaries.

      I attempted to create a CER community based upon the DARPA model. We held a two-day CER principal investigator (PI) meeting⁹³ in February 1984 where the PIs presented their research in a series of focused sessions. Even though large, integrated projects usually were not the primary focus, the new state-of-the-art facilities resulted in many significant research projects. Jack Schwartz’s 1983 taxonomy of parallel computers⁹⁴ included several that were developed or extended under CER grants. These included the NYU Ultracomputer,⁹⁵ the Illinois CEDAR machine,⁹⁶ the Texas Reconfigurable Array Computer (TRAC),⁹⁷ the Berkeley Hypertree (also at Wisconsin),⁹⁸ the Utah Applicative Multi-Processing System,⁹⁹ the Wisconsin GAMMA database machine,¹⁰⁰ the Maryland ZMOB,¹⁰¹ Yale’s ELI-512 computer,¹⁰² the Duke Boolean Vector Machine,¹⁰³ and the Blue CHip Project¹⁰⁴ at Washington (begun as the Purdue Configurable, Highly Parallel (CHiP) family). The Eden Project¹⁰⁵ expanded on ideas from the efforts at Xerox PARC, SRI, and other industry labs, and developed an influential operating system. In a similar direction, the Crystal Project¹⁰⁶ at Wisconsin developed a shared multicomputer. The Cornell CER started a long career by Ken Birman¹⁰⁷ in distributed operating systems, which included the Isis Toolkit, the Horus system, the Ensemble system, and currently Isis2, Gradient, and the reliable TCP solutions. At Cornell, Bob Constable worked with Birman on Horus and Ensemble and developed a program development system called PRL (“pearl”)¹⁰⁸ that provides automated assistance with explaining and proving. There are many additional examples.

      Several CER grants became the basis for early Science and Technology Centers relating to computing: the $38 million Science and Technology Center for Research on Parallel Computation at Rice University with the California Institute of Technology, Syracuse University, the University of Tennessee, Argonne National Laboratory, and Los Alamos National Laboratory (NSF 9120008); the $21 million Center for Research in Cognitive Science at the University of Pennsylvania (NSF 8920230); and the $35 million Science and Technology Research Center in Computer Graphics and Scientific Visualization at the University of Utah with Cornell University, Brown University, the University of North Carolina, and the California Institute of Technology (NSF 89202191). The more recent Team for Research in Ubiquitous Secure Technology (TRUST) at Berkeley, with Carnegie Mellon, Cornell, Mills College, San Jose State, Smith College, Stanford, and Vanderbilt (NSF 0424422) can trace some of its activities back to research that came out of the Cornell CER СКАЧАТЬ