Computing and the National Science Foundation, 1950-2016. William Aspray
Читать онлайн книгу.
was concern early in the CER program that these investments were severely limiting the funds available for regular grants. Jim Ortega at the May 29, 1981, CS Advisory Committee requested a review of the impacts of CER and CSNET on Standard Research Project Support (SRPS). While the Computer Science Section budget had increased 24% from FY 1980 to FY 1982, SRPS support had only increased 10.4%. In comparison, the Mathematical and Physical Sciences budget increased 19%. After substantial discussion, the advisory committee concluded that “the CER and CSNET are essential to the furtherance of computer science research and that it is too early to modify the direction being taken.”109
In 1982, the DoD planned to expand its agencies’ support to include as many as 10 or 15 institutions. This DoD program never materialized, but DARPA upgraded facilities for its major contractors and expanded its smaller ($250–300,000) equipment contracts. ONR was able to provide a few Special Research Opportunities contracts in computer research with some facilities support. Without the planned DoD programs, the CER program grew in an attempt to fill the need.110 Through 1985, NSF had committed $49.89 million to 22 institutions for experimental computer research. In addition, DARPA had major contracts with MIT, Stanford, Carnegie Mellon, and California-Berkeley, which had supported experimental computer research. When NSF began the CER activity, it expected to support approximately 15 institutions. With more than 70 Ph.D.-granting departments of computer science and engineering, it was estimated that 25 to 30 would require research facilities of the magnitude provided by the CER program.
A report in 1986111 noted that “universities have been funding CS growth at rates significantly higher than in any other major discipline. But national funding policy has favored the growth of basic research in CS at a rate no greater than that of other scientific, mathematical, and engineering disciplines.” The authors warned that “the late 1980s will witness the departure of our best and most mobile computer scientists and graduate students for industrial careers. Inevitably, the universities will be unable to maintain the centers of academic excellence in CS that have been so carefully developed during the past five years.” A year later, “[t]here has been a dramatic increase in federal funding for both total and academic CS research between FY 1976 and FY 1987 . . . Funding has shifted away from basic and toward applied research, both in CS federal funding as a whole and within academic CS.”112 NSF convened an Infrastructure Workshop in July 1991. The workshop report113 placed a high priority on maintaining the Institutional Infrastructure programs at $20 million per year. It also proposed developing a matching program of $8 million to support facilities for individual and small group grants.
Recognizing that no more than 30 computer science departments114 would have enough experimental computer scientists to require CER-scale funding, a new Institutional Infrastructure program was announced with both “Large-Scale” (II-LS) and “Small-Scale” (II-SS) grants. Given the shortcoming of DoD funding, described above, CISE invested in, expanded, and replenished the experimental facilities at around 30 institutions. The II-LS program essentially replaced CER. II-SS was aimed at units with fewer experimental computer scientists and a reduced need for facilities support. Figure 2.3 shows both “large” (CER and II-LS) and “small” (II-SS).
The CISE Institutional Infrastructure program continued until 1993, when it was replaced by the CISE Research Infrastructure (RI) program. The RI program had institutional, instrumentation, and “shared” facilities, such as the CISE Advanced Distributed Resources for Experiments (CADRE). Figure 2.3 shows that 60 institutions benefited from the CER, II-LS, II-SS, and RI—many receiving three or more awards. In 1989, CISE introduced a facilities program directed toward minority-serving institutions (see Figure 2.4). One of those awards to University of Texas at El Paso became the basis for the CISE BPC CAHSI (Computing Alliance of Hispanic-Serving Institutions) Alliance, and in turn the CAHSI INCLUDES project, one of the first five $10 million NSF INCLUDES Alliances.
When I returned to NSF in 2000 as Division Director for Experimental and Integrative Activities, we moved the Minority Institutional Infrastructure to the education and workforce block of programs and began to redesign the remaining infrastructure programs of CISE. Today, CISE supports a Community Research Infrastructure (CCRI) program to encourage “discovery and learning in the core CISE disciplines . . . by funding the creation and enhancement of world-class research infrastructure. This research infrastructure will specifically support diverse communities of CISE researchers pursuing focused research agendas in computer and information science and engineering.”115
Figure 2.3Infrastructure awards FY 1980–1998.
Figure 2.4Minority institutional infrastructure awards FY 1989–1998.
2.10CSNET
In parallel with the “crisis” in experimental computer science, a number of researchers at leading universities did not have access to the ARPANET.116 Curtis and Pasta’s strategy for the Coordinated Experimental Research program included a computer network for research. Lawrence Landweber invited a number of researchers and government representatives to the University of Wisconsin–Madison in May 1979. His goal was to “discuss how computer network services like those of ARPANET could become available to the entire community of computer science researchers.”117 The attendees included Kent Curtis, Bob Kahn, and individuals who had experience with Theorynet and other similar “mailbox” systems hosted on commercial networks.118 The participants agreed that ARPANET’s mail, file transfer, and remote login services had “enhanced research productivity and had generated a strong community spirit among computer science and engineering departments that hosted ARPANET sites.”119
A consortium of universities including Georgia Tech, Minnesota, New Mexico, Oklahoma, Purdue, UC-Berkeley, Utah, Virginia, Washington, Wisconsin, and Yale submitted a proposal in November 1979 for a “CSNET” that would create a separate and independent network to provide ARPANET-like services to all U.S. computer science departments. Given the cost of duplicating the ARPANET infrastructure (estimated at $100,000 per institution), the proposed network would be built on commercial X.25 networks such as Compuserve, Tymnet, and Telenet. NSF declined the proposal in March 1980. Reviewers felt that the proposers were reinventing the ARPANET and not extending it, that they lacked a strong project management plan, and that for NSF to pay for the network it would have to reduce research support.120 The reviewers’ skepticism was not unlike the reaction I had heard during CER site visits in 1980–1981 when asking proposing PIs about the CSNET plans. Many of those outside the CSNET proposal development did not see a real justification for an ARPA-like network, and some not even the need for email.
The NSF offered to fund a thorough study of CSNET. Landweber organized a meeting in Berkeley on June 15, 1980, at which DARPA announced its support for CSNET and assigned Vinton Cerf to help develop a plan to connect CSNET and the ARPANET. Landweber convened a 19-person CSNET planning committee, including Cerf and others who had extensive computer networking experience. The group worked throughout the summer of 1980 to devise an implementation strategy. The outcome was a plan to design CSNET as a network on multiple communication platforms interconnected via an Internet protocol. DARPA was moving from NCP to TCP/IP and the MMDF-based121 Phonenet system had been developed by David Farber and David Crocker at the University of Delaware. Phonenet was a low-cost mail relay system similar to the UUCP-based mail relay developed by Bell Laboratories to connect computer science departments that had Unix platforms. The UUCP protocol122 supported email and file transfer, but required explicit addressing and, unlike MMDF, was not compatible with IP networks at the time. The proposal would integrate ARPANET access, X.25 networks running TCP, and Phonenet to provide multiple tiers of services and costs for departments wishing to be connected to CSNET.
Landweber and colleagues from