5.3 Growth and Convergence

---

The TCP/IP protocol suite, including DNS, entered the data communication environment at a critical and strategic time. Computers and computer communication were just beginning to diffuse widely throughout the business world. Buyers faced a plethora of competing and incompatible networking standards and protocols, such as IBM's Systems Network Architecture (SNA), the ITU's X.25 protocol, Digital Equipment's DECNET, and a variety of local area network standards. Debates and negotiations over technical standards were complicated by the fact that data communication products and markets cut across a wide swath of industries and interest groups. Computer equipment manufacturers, telecommunication service providers, telecommunication equipment manufacturers, and major users all had a stake in the outcome. Inevitably, national governments viewed standardization negotiations as an extension of their industrial policies, adding political and economic considerations to the debate over technology choices.

During the 1980s the International Standardization Organization led a global effort to develop a standardized, open approach to almost all aspects of data communication. This effort became known as Open Systems Interconnection (OSI). The OSI effort had strong backing from the world's traditional standards bodies, telephone companies, and governments. TCP/IP, on the other hand, failed to gain international political backing outside the United States. And yet, by 1991 or so, it was evident that data communication had begun to converge globally on TCP/IP and Internetstyle domain names. The Internet, not OSI, ultimately became the common ground upon which most networking initiatives met and achieved interoperability. In this book, I am more interested in demonstrating that this happened than in giving a detailed analysis of why and how it happened. But it is worthwhile to spend some time describing the ways in which the rise of the Internet fulfilled some of the key conditions for winning a standards competition.

5.3.1 Critical Mass: Research and Education Networking

The value of a networking standard depends on who else adopts it. No matter how technically advanced and efficient it may be, a communication protocol or piece of equipment is of little use if it is not compatible with one's desired communication partners. Technology adoption choices are thus powerfully shaped by the choices other adopters make. One of the prerequisites of establishing a standard, therefore, is what economists have called critical mass. This is the minimum threshold of other committed users required to make the adoption of a particular technology attractive to a given user. Networks that achieve critical mass can generate selfsustaining growth. Those that do not achieve critical mass wither and die. One common method of achieving critical mass is to subsidize initial adoption (Rohlfs 1974).

TCP/IP was virtually guaranteed a viable critical mass of initial users by DARPA's willingness to subsidize not only the development of the protocols but also the provision of network services utilizing the protocols. The original ARPANET community created a small but strategic cluster of engineers and scientists strongly committed to TCP/IP and DNS. From 1982 to 1987 this group was expanded to include thousands of users in the federally supported scientific research community.

Parallel to the formation of the ARPA-Internet, the U.S. government had supported several discipline-specific networks, such as Energy Sciences Net, NASA Science Internet, and the National Science Foundation's Computer Science Net (CSNET). From 1983 to 1986 all were linked to the ARPANET backbone using the TCP/IP suite and domain-naming conventions. ^{[27 ]}

In 1983, DARPA also created a $20 million fund to encourage commercial computer vendors to write TCP/IP implementations for their machines. That same year, the protocol was included as the communication kernel for the University of California's popular BSD UNIX software. The Berkeley version of UNIX software was distributed free to universities, thus boosting the dissemination of internetworking capability and ARPANET connectivity (Albitz and Liu 1998, 2).

5.3.2 Converging Networks

The Internet grew not by adding new users, but by interconnecting other networks. Contemporaneous with the rise of the Internet were many networks based on different standards and protocols (Quarterman 1990; Abbate 2000, 200-205). Usenet, for example, used the UUCP protocol to distribute text-based discussion groups, attracting tens of thousands of users around the world by the mid-1980s. BITNET and Fidonet were other networking initiatives, one grounded in academic institutions employing an IBM protocol and the other in IBM-compatible PCs using dialup bulletin boards. Both brought thousands of people into computer networking. Other governments also began to support research and education networks in the 1980s (JANET in the U.K., GARR in Italy, the Korean National Computer Network, and JUNET in Japan). Probably the most important form of bottom-up networking was the local area network (LAN), using Ethernet and other standards.

In any competition for preeminence with alternative technologies, the Internet enjoyed a decisive advantage. The TCP/IP suite had been designed from the very beginning to interconnect networks using different, potentially incompatible protocols. The design assumptions underlying TCP/IP projected a world of thousands of heterogeneous, independently administered networks that needed to interoperate. This differed radically from alternative protocols such as X.25, which were based on the assumption that data communication would be dominated by a limited number of public data networks run by telephone companies.

Moreover, the ARPA research community aggressively developed gateways that allowed other networking protocols to communicate efficiently with networks using TCP/IP. As email capabilities became popular, for example, pressure grew to interconnect different email systems. Email gateways played a big role in promoting the spread of DNS as a naming convention. Networks with different naming schemes could register a domain name, and the resource records for the registration could point to a computer on the Internet that would act as a mail forwarder (RFC 1168, July 1990). Eventually the DNS became a common addressing syntax that allowed different networks to exchange email effectively (Frey and Adams 1990, 12-13). BITNET and UUCP began to use domain names in 1986; Fidonet followed in 1988.

5.3.3 The National Science Foundation Backbone

Another major step toward critical mass came in 1986, when the U.S. National Science Foundation (NSF) decided to expand Internet connectivity to the entire U.S. higher education community. Working through ' cooperative agreements' instead of costly procurement contracts, NSF began to support an Internet backbone starting in 1987, the NSFNET. NSFNET was a virtual private network supplied by a partnership of Merit Networks, Inc. (a nonprofit consortium of university computer centers in the state of Michigan), IBM Corporation, and MCI Telecommunications under a cooperative agreement award from the National Science Foundation. ^{[28 ]}The project signaled a transition from a very limited military and research role to broader education-oriented support for networking.

Technically, the NSF backbone had a special role in the Internet hierarchy: it acted as a generic transit, routing, and switching network (NAS 1994, 239). NSF also provided subsidies to 'mid-level networks' that operated regional facilities to carry research and education traffic from universities and other eligible institutions to the backbone. During this period the Internet was not available to the general public, and it was accessible to businesses only under special conditions. The NSF-imposed 'acceptable use policy' (AUP) limited service 'to support [of] open research and education in and among U.S. research and instructional institutions, plus research arms of for-profit firms when engaged in open scholarly communication and research.' ^{[29 ]}Still, the TCP/IP Internet was achieving a critical mass of users large enough to put palpable pressure on research and education networks in other parts of the world to become compatible with it. Table 5.1 shows the growth in the number of hosts connected to the Internet in the first decade after the formal specification of the protocols.

Table 5.1: Number of Internet Hosts

Date	# Hosts
August 1981	213
May 1982	235
August 1983	562
October 1984	1,024
October 1985	1,961
November 1986	5,089
December 1987	28,174
October 1988	56,000
October 1989	159,000
October 1990	313,000
October 1991	617,000

Source: RFC 1296 (January 1992)

Though they were not part of the original ARPANET cadre, those who rose to prominence in research and education networking at this time became leaders in the broader Internet technical community. David Farber at the University of Delaware and Lawrence Landweber at the University of Wisconsin, founders of CSNET, later became board members of the Internet Society. BITNET and CSNET merged in 1989 under the Corporation for Research and Education Networking, which was run by Mike Roberts, the director of Educom. Roberts also helped to found the Internet Society and later became the first president of ICANN.

As the civilian Internet grew, both the Internet technical community and civilian federal government staff members knew that something important and valuable was happening. 'We were interested in the grand vision,' one NSF official put it, 'and it worked.'

5.3.4 Internationalizing Name and Address Assignment

Along with the growth of the Internet came pressures to distribute internationally parts of the name and number assignment functions.

Local area networks were spreading throughout Europe, and scientists in fields such as physics and computer science often used Berkeley UNIX, which came equipped with TCP/IP for LANs. As LANs were connected into European wide-area networks using TCP/IP, a need for name and number coordination and other forms of cooperation arose. This was the rationale behind the formation of RÈseaux IP EuropÈens (RIPE) in Amsterdam in 1989.^[30] Like many Internet-related organizations, it began as a volunteer effort. The RIPE name was chosen deliberately to tease the European Community funders, whose OSI initiative was named RARE.

RIPE was in a delicate position. European governments backed OSI over Internet standards, and many of the participants were engaged in publicly funded research. ^{[31 ]}One of the common arguments used by the OSI camp against the Internet was that IP addresses were too scarce, and their allocation was under the control of the U.S. government. Members of RIPE convinced the Americans that these complaints could be countered if some part of the address space was delegated to Europe so that locals would control their assignment. ^{[32 ]}The Americans knew that many networks around the world were joining, or attempting to join, the Internet. But the locus of administrative power within a single national government was becoming an obstacle to this growth.

The National Science Foundation's 'acceptable use policy,' for example, tried to prevent commercial use of the NSF-subsidized Internet backbone. The United States enforced the acceptable use policy primarily by controlling entries into the address registration and domain name system databases. Before registrations could be entered in the DDN-NIC databases, thereby enabling global connectivity, a registrant needed a 'sponsor' from a U.S. government agency. Such a policy, however, required foreign networks to adhere to U.S. access and use criteria even if a large portion of their traffic didn't go through the federally sponsored networks.

Beginning in 1990 the Internet technical cadre proposed to make it possible to register a domain name (and in-addr.arpa entries) without a U.S. government sponsor (RFC 1174, August 1990). ^{[33 ]}The first non-U.S. address registry, the RIPE Network Coordination Center (RIPE-NCC), was established in April 1992. A few years later the Internet technical community in the Asia-Pacific region, following the European precedent, established AP-NIC to delegate addresses in that region.

Another critical metric of the internationalization of name and address administration was the delegation of country code top-level domains and the formation of domain name registries overseas. A top-level domain delegation meant that domain name registries capable of assigning and registering second-level domain names were being established in other countries. As the networks in these countries were entered into the DNS root, the networking community in those countries acquired a stake in Internet administration.

The first three country code delegations were made in 1985. From 1986 to 1990 about ten were added each year. That pace doubled from 1991 to 1993 (table 5.2).

Table 5.2: Number of Country Code Delegations, 1985-1993

Year	No. of country code delegations
1985	3
1986	10
1987	19
1988	27
1989	35
1990	46
1991	68
1992	85
1993	108

Source: France-Network Information Center (FR-NIC)

It was Postel at ISI who filled the critical role of assigning country code top-level domains to specific applicants. From 1985 to 1993, Postel made these delegations using a commonsense application of the DNS concept of a 'responsible person.' Delegations were made on a first-come/first-served basis as long as certain basic administrative criteria were met. The administrative contact, for example, was expected to be located in the actual territory that the code symbolized. Significantly, that delegation method tended to bypass completely the institutions in other countries that historically had possessed authority over communication, such as government ministries or post, telephone, and telegraph monopolies. Almost none of these institutions were paying attention to the Internet at this time. Typically, the result was that delegations ended up in the hands of university computer science departments or education and research networking organizations in the named territory. Postel himself noted that the person in charge of assigning second-level domain names 'is generally the first person that asks for the job (and is somehow considered a ‘responsible person').' ^{[34 ]}

There was at this time no explicit policy for resolving competing applications for the same assignment. When conflicts occurred-and they began to after 1991 as governments opened up the Internet service provider market to commercial entry-Postel typically used subtle forms of pressure to prod the disputing parties to settle it among themselves, such as refusing to make any delegation until the disputants agreed on a solution. ^{[35 ]}Contention over country code delegations was one of the first indications of the growing value of top-level domains and the ensuing political and economic obligations of managing the root.

^{[27 ]}Interview with Charles Brownstein, former NSF division chief, June 2000. See also NAS (1994), 238.

^{[28 ]}NSFNET cooperative agreement citation.

^{[29 ]}NSFNET Backbone Services Acceptable Use Policy, June 1992, <http:// www.merit.edu/merit/archive/nsfnet/acceptable.use.policy >.

^[30]RIPE Terms of Reference, November 29, 1989, <http://www.ripe.net/ripe/docs/ripe-001.html>.

^{[31 ]}Notes to

^{[32 ]}Interview with Daniel Karrenberg, June 20, 2000.

^{[33 ]}Interview with David Conrad, August 23, 2000.

^{[34 ]}The proposal was formatted as a recommendation from the Internet Activities Board to the Federal Networking Council and was eventually approved. See sections 5.4.1 and 5.5.1 for more detail about the institutional environment in which these decisions were made.

^{[35 ]}Jon Postel to msggroup, November 15, 1985.

---