THE NSF: ‘MIDWIFE’ TO THE INTERNET

When it comes to networks, success breeds transitions. For, just as the growing value of the telephone network gave rise to a diversified competitive market, so too did the increasing externalities and added value associated with ARPANET generate private­sector interest in developing for-profit data communication services (Cave and Mason, 2001; Rohlfs, 2001).

In January 1981 NSF approved a proposal from Professor Lawrence Landweber of the University of Wisconsin to provide five years of funding to develop a computer science network linking university campuses across the United States. This proposal reflected growing demands by university computer scientists to gain ARPANET access, which was then only open to defense-related researchers (Rogers, 1993, p. 214; Ryan, 2010, p.

In spite of the proposal’s merits, the NSF had some serious concerns, which led to the rejection of Landweber’s first request made in 1979. Not the least of these was the issue of management. Focusing on peer-reviewed basic research, the NSF was ill equipped to manage a large-scale infrastructure project (Rogers, 1993, p. 215). Continued interest in the burgeoning field of computer science led to a second look, however. With the encour­agement of the computer science staff within the NSF, Landweber’s second proposal was approved in January 1981. Accordingly, the NSF pledged to create a Computer Science Network (CSNET) with support for five years, after which it would be self-supporting. Backbone users were required to conform to NSF’s ‘Acceptable Use Policy’ restricting traffic to open research and education and prohibiting commercial activity. Usage fees ranged from $5000 annually for universities to $30 000 for industry facilities (Ryan, 2010, p. 94).

Moving from CSNET to the NSFNET required more than financing; it required buy-in from government agencies and academic science programs. Over the next three years, NSF’s effort gained increasing support due in part to a changed environment. Just as ARPANET had been inspired by Sputnik, the NSFNET gained traction after the Japanese launch of its fifth-generation supercomputing program. Concerned lest the USA lose its pre-eminence in microelectronics and computing, Congress asked the NSF to create a supercomputing program so as to make high-performance computing avail­able to all fields of science (Rogers, 1993, p. 216; Abbate, 1994).

Coordinating and integrating the diverse players needed to carry out the NSF’s mission posed significant challenges.

As a first step, the NSF set out to support supercomputing research at all US universi­ties. To this end, it developed a multi-tiered system, much like the network itself, whereby each level was both independent of, but also interdependent on, the others. Accordingly, each university center was to build its own network with seed money from the NSF. These research centers were then connected at the regional level, using leased lines from local and regional network providers. These networks, of which there were seven by early 1988, constituted the mid-level network. To get them up and running, the NSF initially sub­sidized them, but they were expected to become self-supporting. The regional networks were then connected to the high-speed national ‘backbone’ network, CSNET (Rogers, 1993).

CSNET, however, was poorly equipped to handle the subsequent increase in traffic, or to manage a network of such rapidly growing size (Kazumori, 2003; Ryan, 2010, p. 96). Cost was one issue. In 1992, the NSF spent approximately $11 million annually on NSFNET, and an additional $7 million to support the regional networks. Close to 80 percent of these costs went to leasing lines and routers (MacKie-Mason and Varian, 1994, pp. 5-6). Management costs were also high, as NSF had only 14 staff members to oversee the network (Kazumori, 2003). Hence in 1987, the NSF turned over the construc­tion and management of the backbone network, subsequently renamed the NSFNET, to the Michigan Educational Research Information Triad (MERIT) in collaboration with MCI and IBM at the cost of $14 million (Hart et al., 1992, p.

Under MERIT’s tenure, the NSFNET backbone network underwent a major trans­formation. Employing many of the same concepts and participants as the ARPANET, the NSFNET was deliberately designed to ensure widespread access, interoperability and long-term sustainability (CSTB, 1996, p. 29). To accommodate increased traffic and high-speed users, MERIT upgraded transmission lines from 54 Kbps to speeds ranging from 1.5 to 45 Mbps. Importantly, the NSFNET also employed the TCP/IP rather than the more formal but underdeveloped International Standard Organization’s Open Systems Interconnection (OSI) suite (Rogers, 1993, p. 220).

TCP/IP gave the NSFNET its decentralized, modular, ‘end-to-end’ architecture, whereby application-specific functions were located on the end users’ hosts at the ‘edges’ of the network (Saltzer et al., 1984; Blumenthal and Clark, 2001). This allowed all par­ticipants to manage their own portion of the network, while still being able to connect to the network as a whole. Given this modular design, the NSFNET was very flexible in handling new service providers, innovations, applications, and so on (CSTB, 1996). Changes to the system could be made on a piecemeal basis rather than in a costly system­wide overhaul. In this way, several new protocols were added to TCP/IP to accommodate new services (Hart et al., 1992, p. 673).

This design fostered a critical mass of users, generating a ‘bandwagon effect’. As Jeffrey Rohlfs (2001) has pointed out, interconnection leads to benefits for virtually everyone. A major source of NSFNET’s externalities was the expanding user base. Individuals ben­efited from greater interconnections and the increased applications that additional users inspired (Cave and Mason, 2001, p. 198). There were also a number of complementary externalities, resulting from the advent of personal computers, the declining cost of computing, the development of a Domain Name System, as well as the addition of new search applications such as Gopher, Veronica, and Archie (Abbate, 1994; Rohlfs, 2001; Ryan, 2010). Characterizing these network effects, Rohlfs points out that the NSF, by providing open interconnection at a subsidized cost, ‘internalized the externalities for the greater public good’ (Rohlfs, 2001, p.

The result was precipitous growth in NSFNET traffic. In the fall of 1985, only 2000 computers could access the network. Two years later there were 30 000, increasing to 159 000 over the following two years (Abbate, 1994, p. 180). Internet traffic kept pace. Between May 1989 and May 1991, Internet traffic expanded from 1 billion packets per month to 7.65 billion packets. Two years later it was almost twice as high, at 14.9 billion packets (Hart et al., 1992, p. 671).

These externalities did not go unnoticed. Anticipating the potential profits, MERIT spun off a non-profit corporation in 1990 called Advanced Network Services (ANS). By subcontracting the backbone network’s operations to ANS, MERIT was free to create a for-profit subsidiary - CO+RE - to offer services to the growing number of busi­nesses users of TCP/IP (Ryan, 2010, p. 190). Other organizations, including the recently divested regional Bell operating companies, joined partnerships to realize similar oppor­tunities (Rogers, 1993, p. 220; Hart, 1992, p. 676; Frischmann, 2001). Because commercial networks could not run over NSFNET, these providers had to build their own backbone networks, assuring that the capacity and reach of the backbone was greatly expanded. By the mid-1990s, the regional companies had become national, commercial Internet provid­ers with commercial networks paralleling the NSFNET. Importantly, given the low cost of entry, new players entered the market, bringing with them adaptations of technology for new uses, locations, market settings and applications (Greenstein, 2001, p. 153-4). As a result, a competitive market for high-speed national computer networking services emerged (Frischmann, 2001, p. 18; Ryan, 2010, p. 191).

Signaling a changing narrative and a new kind of rhetoric, the Internet community welcomed commercialization (Rosenzweig, 1998; Ryan, 2010, p. 191). NSF managers also favored these developments, given their limited management capacity and financial resources.

Congress confirmed this approach on 11 September 1991 with the passage of the High-Performance Computing and National Research and Education Network Act of 1991. This Act cancelled the NSF Acceptable Use Policy. At the same time, it reiterated the government’s support for science and for the expansion of the Internet to the public at large though the development of a National Research Education Network (NREN), which, as Rogers points out, became ‘the vehicle for discussing the information infra­structure of the future’ (Rogers, 1993, pp. 223-4).

The NSF encouraged and supported the development of the mid-level regional net­works by fostering network synergies through partnerships of universities, network providers, and businesses, such as BARNET, SURNET, and NYSERNET (ibid., p. 222). These partnerships not only brought diverse providers together, they also helped to aggregate user demand (Bailey, 1995). Notable in this regard was the development by the NSF and MCI of very high-performance Backbone Network Service (vBNS) to provide university communities with broadband capabilities. This effort was continued under the auspices of EDUCAUSE in the post-NSF period in the form of the Internet2 Project, which spawned the Abilene Network.²

Commercial arrangements for interconnection also helped to determine the structure of the emerging Internet field. Acting through its proprietary company CO+RE, MERIT asserted its dominance based on its exclusive ability to move commercial traffic over the NSFNET. Its leverage was significant, given that at the time approximately 35 percent of all network sites could only be reached by travelling on the NSFNET backbone network (Kazumori, 2003). In February 1991, three commercial ISPs, PSINET, UUNET, and CERFnet of California sought to offset this advantage by forming the Commercial Internet Exchange (CIX) with a router in Santa Clara, California (Wheeler and O’Kelly, 1999, p. 328). CIX aimed to interconnect its members on a peering basis, for an initial one-time fee of $10 000.

Concerns about unfair competition led to a Congressional hearing in March 1992. The House Subcommittee on Science called on the NSF to establish four network access points (NAPs), where providers could work out their interconnection agreements, either bilaterally or multilaterally. In June 1992, ANS and CIX agreed to interconnect. The NAPs, each operated by a different telecommunications company, were subsequently created in 1994, located in San Francisco, Chicago, New York and Washington, DC. At each of these ISPs could interconnect and share data in a peering fashion. Generally speaking, the large national ISPs established bilateral peering agreements and negotiated payment schemes with smaller providers. Smaller providers, which might serve only a few thousand customers, established a number of multilateral agreements, allowing them to send traffic on others’ communication lines (ibid.).

While the transition to the commercial Internet went relatively smoothly (Greenstein, 2001), its success was by no means preordained. Building a market not only entails tremendous coordination problems and transaction costs, it also requires building trust, overcoming problems of collective action, establishing interoperability of human practices as well as technology, and creating - either through shared norms or legal requirements - an institutional framework, or as Douglass North (1990) calls it, ‘the rules of the game’. In this regard, the NSF served as a midwife, providing the essential conditions to pave the way for the commercial birth of the Internet. In the years during which the NSF oversaw the NSFNET, it provided an institutional platform to match the Internet architecture, which attracted an ever-increasing number of diverse entities to create a dynamic common-pool resource rife with positive externalities. Just as impor­tantly, the NSF provided a safe, and relatively cost-free site where all the divergent par­ticipants could converge to negotiate the Internet field and work out a governance system to preserve the commons (Ostrom, 1990).

Most notable was the development of Internet norms and governance institutions that sustained themselves even in the context of commercialization (Brown, 2013). These include, for example, the Internet Activities Board (IAB), which was established in 1984 to manage and guide the development of the network and its protocols. (In 1992 it took the name Internet Architecture Board to reflect its increasingly global composition.) Under its purview were two additional organizations: the Internet Engineering Task Force (IETF), formed in 1986, which was charged with addressing short-term, practi­cal issues; and the Internet Research Task Force (IRTF), which was set up in 1989 to examine long-term issues. In addition, the Internet Society (ISOC) was founded in 1992 to serve as an umbrella organization that would provide leadership for the IETF and the IAB (Ryan, 2010, pp. 101-2). Finally, the Internet Corporation for Assigned Names and Numbers (ICANN) was established in 1988 as a non-profit global organization responsi­ble for coordinating the Internet system of unique identifiers.³

26.6