Computing and the National Science Foundation, 1950-2016. William Aspray
of Delaware, Purdue University, RAND Corporation, and the University of Wisconsin submitted a revised CSNET proposal to NSF in October 1980, and the National Science Board (NSB) approved the five-year proposal the following January. To address concerns about how CSNET would be managed required an unusual structure in which NSF itself, under Project Director C. William Kern, would directly manage the project for two years (through 1983) by means of contracts. NSF management would focus on setting up the organization to collect and disburse funds, and after two years the project would be sufficiently advanced that users would be willing to begin paying dues and fees.123 Contracts were established with the University of Delaware, Purdue University, Rand Corporation, and the University of Wisconsin for CSNET development. Bolt, Beranek, and Newman (BBN) was contracted to run the CSNET Coordination and Information Center (CIC) for managing the network and distributing software.
Figure 2.5CSNET architecture 1981.
On March 6, 1981, NSF announced the establishment of CSNET, which would become a major step along the path to the Internet. On May 28, 1981, Bill Kern presented the status of the CSNET effort to the Computer Science Advisory Committee.124 He discussed the two-year NSF management plan and the expectation that CSNET would become self-supporting in five years. He also told the Advisory Committee members that DARPA would develop the CSNET/ARPANET gateway and that software, systems, and services would target the Berkeley UNIX 4.3BSD operating system on VAX computers. He indicated that CSNET would initially comprise three subnets (Figure 2.5)—ARPANET, Telenet, and Phonenet—but would be designed to support expansion to other available networks. CSNET initially provided the same services as ARPANET: mail, file transfer, remote login, and an on-line name server. CSNET’s $5 million project budget, limited staffing, and the five-year timeframe for self-sufficiency put significant pressure on the CSNET team.
Figure 2.6CSNET map 1983.
In just six months, CSNET was operating,125 including a Phonenet site at NSF in the Computer Science Section, the first NSF Internet connection. In addition to NSF, Phonenet sites included Cornell, FCC-NET, HP Labs, Purdue, Princeton, UC Irvine, and Delaware, with plans to expand to New Mexico Tech, Pennsylvania, Georgia Tech, Duke/UNC, Fairchild, and Maryland-College Park.
When Kern stepped down as CSNET Project Manager in October 1982, I assumed the role of CSNET Project Director with Landweber as Chair; Peter Denning, Richard Edmiston, David Farber, Anthony Hearn, Kern, and me as members of the Management Committee. By the time of the first CSNET Newsletter,126 56 Phonenet sites were operational and 27 were nearing operation (see Figure 2.6). These connected through the two CSNET relays on the ARPANET at RAND Corporation and the University of Delaware. CSNET was beginning to meet with European network leaders to investigate international connections.
After its two-year management, the NSF selected the University Consortium for Atmospheric Research (UCAR)127 on May 3, 1983, to host and manage CSNET, with Leonard Romney (UCAR executive director) as PI and a member of the CSNET Management Committee.128 As UCAR assumed control of CSNET, the Management Committee was replaced by a larger Executive Committee129 with Peter Denning as chair, representing the computing research community; and the operation of the CSNET relays and technical services moved entirely to BBN. BBN housed the CSNET Coordination and Information Center (CIC) to provide operational management of CSNET. In 1983, CIC staff included: Dr. Richard Edmiston (CIC Director); Laura Breeden (CIC User Liaison); Dan Long (CIC Technical Liaison); and Beth Johnson (CIC Staff Assistant). Leonard Romney left UCAR in May 1984 and was replaced by Stanley Ruttenberg.
Figure 2.7CSNET executive committee 1983.
By October 1983, Lawrence Landweber was leading an effort to create gateways and connections among BITNET, and Canadian and European networks, including SERCNET (United Kingdom), SUNNET (Sweden), CERNEY (Switzerland), and UNINET (Norway). CSNET connected to BITNET through a University of Wisconsin gateway. At the time, BITNET was a fast-growing network connecting university computing centers via IBM store-and-forward software and leased lines. Connecting to international and other U.S.-based networks raised issues about how to manage the costs associated with traffic transiting multiple networks. In 1983, the initial agreement called for each network to bear the costs of message traffic into other networks. Security issues also arose concerning international traffic in and out of ARPANET via BITNET and CSNET.
In June 1984, I described new NSF networking plans (see NSFNET below) to the CSNET Executive Committee and asked them how CSNET might interact with this expanded vision. CSNET established new gateways with SUNET (Sweden), the Israeli Network, and DFN (Germany). NSF paid for CSNET dues for undergraduate institutions. Dennis Jennings, then chairman of the European Academic Research Network (EARN), visited the CSNET Executive Committee in September 1984. He would soon be recruited to NSF, replacing me as the program director for Networking in the Office of Advanced Scientific Computing (OASC).
The last NSF payment for CSNET operations was in mid-1985. By 1986, CSNET connected more than 165 university, industrial, and government computer research groups serving more than 50,000 researchers and students, including accounts for 1000 Internet hosts. Network services were operational and numerous networks outside the U.S. were connected.130 CSNET was self-supporting and received significant industry funding. CSNET clearly demonstrated, for the first time, that users were willing to pay for network services.
CSNET actively collaborated with colleagues in other countries, supporting and often enabling the international expansion of the Internet. CSNET had mail connection via CSNET/Internet and USENET/EUNET/UUCPNet connections to foreign affiliates and their gateways. These included: CDNNET (Canadian Academic Network, via the University of British Columbia); SDN (System Development Network, with a gateway at the Korea Advanced Institute of Science and Technology); SUNET (Swedish University Network, via Chambers University of Technology); CHUNET (Swiss University Network, via ETH-Zentrum); INRIA (French University Network, through INRIA/Rocquencourt); DFN (Deutches Forschungsnetz); JUNET (Japanese University Network, through the University of Tokyo); Finnish University Network (via Helsinki University); AC.UK (Academic Community, United Kingdom, via University College, London); ACSNET (via a UUCP-based connection at the University of Melbourne); New Zealand Academic Network (via Waikato University, Hamilton); and the Israeli Academic Network (via Hebrew University of Jerusalem).
At its meeting in Ann Arbor in June 1988, the CSNET Executive Committee discussed a potential merger of CSNET131 and BITNET. As vice chair of the Executive Committee, I was assigned to the CSNET-BITNET merger team, planning a merged network called “ONENET.”132 Eventually, in 1989, CSNET and BITNET were brought under the Corporation for Research and Educational Networking (CREN), a non-profit corporation initially composed of the organizations that had participated in BITNET and CSNET. NSF funded the expansion of CSNET and BITNET, as well as the development of TCP/IP services as adjuncts to NSFNET. Because of the success of NSFNET and the regionals, CREN discontinued CSNET services in 1991. CREN ended their support for BITNET in 1996, due to the growth of TCP/IP-based networks, and by 2003, CREN dissolved itself.
2.11The Office of Advanced Scientific Computing and NSFNET
Beyond a brief overview of the high-performance computing programs, the details of which are covered in Chapter 10, this subsection examines the developments that led to the NSFNET. Chapter 9 provides details on NSF’s broader role in networking before, during, and after the NSFNET project.
The Lax Report133 identified two problems: “important segments of the research and defense communities lack effective access to supercomputers and students are neither familiar with their