Jennings identified several key decisions made under his leadership.149 The first was to develop a general-purpose network for all science and engineering research rather than a network only providing supercomputer access. There was considerable disagreement on this issue between the OASC networking subcommittee and John Connolly—and to some extent the centers. Connolly was reluctant to separate the network development from the centers, but Gordon Bell eventually split the networking program off as a separate division in the Computer and Information Science and Engineering (CISE) Directorate, as described below.
Another important decision was to adopt a “network of networks” approach. CSNET employed a network of networks approach by integrating Phonenet dial-up services, public network X.25 services, and ARPANET; and ARPANET, as it transitioned to an R&D network, was also integrating networks with quite different communications layers: satellite, phone lines, etc. After a visit to the Cornell Theory Center, Jennings met with Richard Mandelbaum, who was then working with Cornell and other New York state universities, corporations, and research laboratories to develop a statewide network, NYSERnet. Jennings provided some seed funding to NYSERnet, and later to other regional networks including SURAnet, BARRnet, MIDnet, Westnet, Merit, NorthWestNet, and NEARnet. The network of networks model evolved from supporting networks150 with differing transport and physical layers and a common Internet layer (such as in CSNET) to also support “tiered” networks that included campus Local Area Networks (LANs), regional networks, and a national backbone. Similarly, the NSFNET program funded the center-based SDSC and the JVNC networks.
As an interim arrangement in October 1985, NSF and DARPA151 agreed that ARPANET could be used to access the centers hosted on ARPANET (Illinois, Cornell, Minnesota, and Purdue). This agreement opened up ARPANET hosts, typically servers in computer science and engineering departments, to a broader set of users via campus-wide networks. The NSFNET program also funded CSNET and BITNET to develop advanced TCP/IP services to provide similar access.
In September 1985, NSF announced its intention to implement a national backbone linking the five NSF supercomputer centers and NCAR, with connections to regional and campus networks. There was some initial pushback from the centers, concerned that they could lose customers in moving from “star networks” and proprietary protocols to a broadly accessible national network with common protocols. A related decision was the selection of Dave Mill’s “fuzz-ball” PDP-11-based routers due to the high cost of ARPANET Interface Message Processors and the lack of commercial alternatives.
The decision for which Jennings may be best known is the adoption of the DoD TCP/IP and related ARPANET protocols as the standard for NSFNET. NSF had originally intended to use the International Standards Organization (ISO) Open Systems Interconnection (OSI) protocols, but they were not yet widely available.152 The scientific communities planning to use the centers had developed preferences for protocols used by specific disciplines: MFENET by the magnetic fusion energy community, DECNET by the high energy physics community, etc. TCP/IP was mostly available on Unix systems and not on the Cray Time-Sharing System (CTSS) that was running at many of the centers.
Jennings left NSF at the end of March 1986, having developed a model for NSFNET and moving it forward. He had established a Networking Technology Advisory Group (NTAG) and put a staff in place. Following a brief stint as acting president at the John von Neumann Center, he returned to Ireland and was replaced at NSF by Steven Wolff. Wolff had met Dave Farber when DARPA was arranging an ARPANET connection for the Delaware CSNET Relay. Wolff was working at the Army Ballistic Research Labs (BRL) located in the Aberdeen Proving Ground in Maryland, and BRL had provided the ARPANET line connecting the University of Delaware for the CSNET relay. Farber convinced Wolff to join NSF on a detail from BRL as NSFNET program director. Wolff brought substantial experience to the position, having served on the faculty of Johns Hopkins after receiving a Ph.D. from Princeton. His experience at BRL had included work on TCP/IP. He became Division Director for the Networking and Communications Research and Infrastructure when Gordon Bell split it off from the Division of Advanced Scientific Computing in April 1985—April Fool’s Day as Wolff153 remembers it. He was responsible for much of the development, expansion, and eventual privatization of NSFNET. Details are in Chapter 9.
2.12The Beginning of CISE
In 1986, the NSF programs and offices supporting computing and information research and applications were brought together for the first time since NSF was founded. The new Directorate for Computer and Information Science and Engineering (CISE) would become the organizational core NSF used to exert federal leadership in computing.
At the request of Richard Nicolson, AD/MPS, in April 1985 Kent Curtis carried out an analysis of the options for organizing computing programs within NSF. Curtis considered the NSF Office of Information Services (OIS), the IT support organization led by Connie McLindon. He concluded that, while OIS was funding some projects such as EXPRES and working with the networking program, it “had no primary research role” and should remain an administrative unit. In his analysis, he looked at programs funding “informatics” viewed broadly. These included the Computer Engineering program, the Division of Computer Research, the Division of Information Science and Technology, the Office of Advanced Scientific Computing, and various elements of Materials Research, Mathematical Sciences, Electrical and Computer Engineering and Behavioral and Neural Sciences. Curtis considered various combinations of these programs and even a new directorate encompassing Mathematical Sciences, Cognitive Science, Linguistics, Systems Engineering, and Management Science. There were “substantial benefits and faults to be expected from any decision” he noted, and added that “the Director should feel free to follow his instincts because there is no obvious wisdom to suggest a particular course.”154
In the late fall of 1985 after I returned to Washington, I began to work with Chuck Brownstein on Bloch’s plans to consolidate NSF computing activities. Bloch officially announced his intentions to create a new directorate and hire Gordon Bell to run it on March 3, 1986. Bell had already begun consulting with Bloch and Mary Clutter. In February, Bell requested that Brownstein take on the role of Executive Officer of the new directorate, Jerry Daen be added as Planning and Administrative Officer, and I join half-time on loan from DCR.
Albert Bridgewater, the Deputy Assistant Director for Astronomical, Atmospheric, Earth, and Ocean Sciences (AAEO), wrote a memorandum155 to Bell in February 1986 suggesting a process and schedule for a new Computer and Information Science and Engineering (CISE) Directorate. This process included meetings with the National Science Board (NSB), developing long-range plans, and presentations to Bloch and the NSB in the spring of 1986. Bell, Brownstein, Daen, and I began to develop a strategy to address the deliverables outlined in the Bridgewater memorandum.
In a second memorandum,156 Bridgewater encouraged Bell to be a proactive Assistant Director Designate by: Identifying areas needing greater emphasis or support; organizing community support; organizing National Academy studies; encouraging links and cooperation with other agencies and directorates; and keeping the National Science Board, the Office of Management and Budget, the Office of Science and Technology Policy, and NSF management informed. Bridgewater was really telling Bell that, with Bloch’s backing and his national credentials, he had an opportunity that had not been given to the NSF computing program leaders in the past.
In Bell’s typed and hand-annotated notes of February 26, 1986, he began to sketch out ideas for CISE:
CISE encompasses fields that are predominately concerned (measured either in design effort or system cost) with the understanding (computer science) and design of computers (computer engineering). These SYSTEMS include: traditional and specialized computers, all forms of computer and communications networks, various transducer interfaces for computers and robots, specialized signal processors, and VLSI circuits and their design systems to implement the particular information processing system.
CISE is not concerned with the phenomena or processes necessary to implement the above systems . . . although it is concerned with the design and implementation of the large systems