5.2 The Origin of the Root

---

The demand for research on internetworking followed quickly on ARPANET's heels. By 1973 the military agency, now named DARPA, was supporting two other packet-based networks (Abbate 2000, 121). Each of these networks used different, incompatible protocols. The military wanted to retain the advantages of specialized networks, but it wanted universal communication among them. It needed an internetworking protocol.

The task of developing such a protocol was taken up by Robert Kahn, who had moved from BBN to DARPA as a program officer in 1972. Kahn conferred with Vint Cerf, who had graduated and moved to Stanford University. In the spring and summer of 1973, Kahn worked with Cerf and others to develop a 'universal host protocol' and common address space that could be used to tie together separate data networks. A basic architecture for a Transport Control Protocol was written by Kahn and Cerf in 1973 and published in 1974.^[5] From 1975 to 1977 various versions of the proposed protocol were implemented in software and tested at BBN, University College London, and Stanford. It was during this period that David Clark, a computer scientist at the Massachusetts Institute of Technology (MIT), became involved.

The TCP was not yet ripe for full implementation. A key breakthrough came in 1978, when Cerf, Postel, and Danny Cohen proposed to split the protocol into two parts. A separate, connectionless Internet Protocol would be used to move packets between machines; a connection-oriented Transport Control Protocol would organize communications between hosts in an end-to-end fashion. With this basic conceptual issue addressed, intense work on a formal implementation proceeded. In September 1981, RFC 791 was presented to DARPA as the official specification of Internet Protocol. ^{[6 ]}The Internet address space had been created.

5.2.1 The First Address Assignments

RFC 790, released at the same time as RFC 791, documented the first IP address assignments to particular organizations. Forty-three class A network addresses were given out at that time. Most went to the local packet radio networks, satellite networks, and other ARPA-supported networks to be encompassed by the internetworking project. But local networks of universities (MIT, Stanford), research organizations (MITRE, SRI), and a few commercial carriers (Comsat, Tymnet, DECNet) also received assignments. Included, too, were several non-U.S. entities, such as the British Post Office, the French Cyclades network, University College London, and the British Royal Signals and Radar Establishments.

Jon Postel was listed as the author of both RFCs 790 and 791. During the formulation of the Internet protocols in the 1977-1981 period, Postel gained recognition as the person responsible for address and number assignments within the small DARPA community. RFC 791 states, with the informality typical of the early RFCs, 'The assignment of numbers is . . . handled by Jon. If you are developing a protocol or application that will require the use of a link, socket, port, protocol, or network number, please contact Jon to receive an assignment.'

After completing graduate school at UCLA in 1973, Postel had moved to the MITRE Corporation (1973-1974), then to SRI International in Northern California (1974-1977), and in 1977 to ISI, where he stayed until his death in 1998. By October 1983 documents indicated that the responsibility for day-to-day assignment tasks had been delegated to Postel's ISI colleague Joyce Reynolds. ^{[7 ]}

5.2.2 The Invention of the Domain Name System

The spiral of growth began almost immediately after the successful implementation of internetworking among the military research networks. The Internet pioneers became aware that all planning had to take continuous increases in the scope of the network into account. The ARPA-Internet in 1982 consisted of only 25 networks and about 250 hosts. MIT's Dave Clark warned in RFC 814 (July 1982) that 'any implementation undertaken now should be based on an assumption of a much larger Internet.' ^{[8 ]}One of the weakest links in this regard was the ARPANET's approach to naming computers. As noted in chapter 3, up to this point the network had translated names into addresses by having each host store a specially formatted table, hosts.txt, containing the name and address of every computer on the network. The authoritative hosts.txt file was maintained on a Hostname Server by the Network Information Center of the Defense Data Network (DDN-NIC), operated by the private company, Stanford Research Institute (SRI) in Menlo Park, California, under contract to the Defense Communications Agency.

As early as September 1981, David Mills of Comsat noted, 'In the long run, it will not be practicable for every internet host to include all internet hosts in its name-address tables' (RFC 799). Some concept of the domain name system was already hatching within the ARPA-Internet community. Mills wrote of a 'hierarchical name-space partitioning,' while Clark in RFC 814 mentioned plans to create a 'distributed approach in which each network (or group of networks) is responsible for maintaining its own names and providing a ‘name server' to translate between the names and the addresses in that network.' ^{[9 ]}The basic concepts underlying the domain name system (DNS) were published only a month later by Su and Postel (RFC 819, August 1982). More detailed specifications and some early implementation software were written by Paul Mockapetris, who also was working at ISI (RFCs 882 and 883, November 1983).

Electronic mail was one of the first networking applications developed by the ARPANET community. The real push for network growth came not from the need to share mainframes but from email, which presented an opportunity to exchange ideas and gather comments from peers. At least since 1975, ARPANET participants had begun to deploy this newly developed capability to create virtual communities capable of collaborating on the development of protocols and standards. One of the first such email lists, if not the first, was the msggroup list moderated by Einar Stefferud. ^{[10 ]}

The implementation of DNS, which took place over the next six years, put to work the techniques for virtual collaboration by email list. In 1983, Postel inaugurated the namedroppers mailing list, 'to be used for discussion of the concepts, principles, design, and implementation of the domain style names.' ^{[11 ]}The group was used to review the documents describing DNS and discuss implementation of the system. Postel had already developed a transition plan. The first top-level domain of the DNS implementation was to be .arpa, a 'temporary' top-level domain. All the names in hosts.txt would take the form hostname.arpa (RFC 881, November 1983). The next step was to define new top-level domains.

5.2.3 Top-Level Domain Controversies

That step (defining top-level domains) proved controversial. Indeed, the criteria for creating top-level domains and the semantics associated with them immediately raised many of the issues that later made domain name conflicts the catalyst of international institutional change.

DNS was organized around the principle that a 'responsible person' would be delegated the authority to assign and resolve names at any level of the hierarchy. Names at all levels were conceived as names for network resources-primarily host computers-not for people, organizations, documents, or products. The designers of DNS had a good idea who they expected to take responsibility for second- and third-level names.

Secondlevel domain names were thought of as names for major organizations whose networks contained 50-100 hosts. ^{[12 ]}Third-level domains would be administered by divisions of those organizations, or by organizations with only one host. The fourth level would be smaller subdivisions of the organizations. The DNS was anticipated to be 'deeply hierarchical' (Klensin 2000).

If the second level consisted of organization names, the top level had to be broader categories or groupings of organizations. What then should those categories be? ^{[13 ]}Who was the appropriate 'responsible person' for them? That issue was hotly debated on the email lists.

In a draft memo issued May 1984, Postel proposed six initial top-level domains: .arpa, .ddn, .gov, .edu, .cor, and .pub. ^{[14 ]}Einar Stefferud immediately voiced a deep criticism of the whole proposal: 'It seems to me that this new draft has gotten us into the troublesome turf of semantic definitions, wherein we attempt to carve up the world and assign responsibility and authority to non-existent entities for large, ill-defined clusters of users and their service hosts.' ^{[15 ]}Inadvertently corroborating Stefferud's argument, another list participant complained, 'I have yet to run into ANYONE

outside the United States who is interested in the EDU/COM/GOV domains. Without exception, they all want the top-level domains to be based on geography and international boundaries.' ^{[16 ]}The British quickly expressed a desire to use a country designator rather than one of Postel's proposed names, although there was disagreement over whether to use .gb or .uk. Eventually .uk was assigned to Andrew McDowell of University College London-the first country code delegation.

Postel was interested in the design and implementation of the DNS, not semantics. Sensing the annoyances inherent in taking responsibility for naming political entities such as nations, Postel looked for an established, fixed list of country names. He found just what he was looking for in a recent standard issued by the International Standardization Organization, 'Codes for the Representation of Names of Countries,' designated as standard ISO-3166. The list, developed to guide interchanges among national postal, transport, and communication authorities, assigned twoletter alphabetic codes to countries and territories. (Unfortunately, the official designation for Great Britain under this standard was .gb, not .uk, but it was too late to alter Postel's original assignment.) In the final version of RFC 920, issued in October 1984, the ISO-3166 list was incorporated as a set of top-level domains. ^{[17 ]}

Reflecting the still unsettled criteria for selecting top-level domains, RFC 920 also authorized a category of the top-level domain that Postel referred to as a 'multi-organization,' a catch-all that would include large clusters of organizations that were international in scope and did not fit into any of the other categories. It seems to have been Postel's response to pressures on the Internet community to give other data networks their own top-level domains.

The DNS-inspired need to impose categories on the networking world raised other controversies as well. Some commentators criticized the proposed top-level domain names because they might confuse users as to which domain a particular organization could be found under. Would Stanford Research Institute be sri.edu, sri.cor, or sri.org? An exasperated Postel replied, 'This is a naming system, not a general directory assistance system.' It was not the job of DNS, he argued, to make domain names guessable by creating unique and intuitive assignments at the top level. 'The whole point of domains, he wrote, 'is to subdivide the name assignment problem. To try to preserve some higher-level uniqueness would require the very central coordination we are trying to eliminate!' ^{[18 ]}Similar assumptions about guessable names, however, played a big role in later debates over trademarks and domain names, and in resolving that problem, the ICANN Uniform Dispute Resolution Policy, as Postel predicted, had to recentralize authority over second-level name assignments (see chapter 9).

Some of the later problems associated with the delegation of Internet country code top-level domains also were dimly anticipated at this time. On the msggroup list, Postel drew a contrast between the Internet world and the more formal and regulated X.400^[19] naming conventions under development by a committee of the International Standardization Organization (ISO) and the International Telecommunication Union (ITU). Postel noted that domain names were generally delegated to 'the first [ responsible] person who asks for the job,' whereas the ISO/ITU approach was to give assignment authority to 'some bureaucrat that does not really want to do it, but is assigned the job by the government-run PTT.' ^{[20 ]}A more pessimistic perspective on this topic was voiced by Stefferud: 'The real domain authorities,' he wrote, 'are going to be selected by some political processes that are not identified well enough, in any of the drafts we have considered, to allow us to seriously consider deciding on any of the TOP-level domains, ARPA included. . . . There just ain't no way that us techies are going to be allowed to dictate domain structures beyond the current bounds of the ARPA and DDN sub-nets.' Stefferud proved to be wrong in the short term: the Internet community was able to define its own domainnaming structures before the 'political processes' he feared caught up with it. But the comment was prescient (see chapters 9 and 11 on the political controversies that emerged over the delegation of country codes).

It was apparent that the meaningfulness of name assignments under DNS had opened up a new world of policy issues. As Steve Kille complained on the namedroppers list, 'Eternal arguments about what everyone is called . . . [have] already filled far more network bandwidth than any of the design discussions.' ^{[21 ]}Even so, the policy debates at this time were not animated by the possibility of economic gains or losses. Once commerce in names entered the picture, these latent controversies became explosive.

Beginning in November 1985, Postel banished semantic issues from the namedroppers list altogether, directing them to msggroup. ^{[22 ]}In 1987 he formally revised the namedroppers list charter to prohibit debates over semantics. ^{[23 ]}

5.2.4 Early Administrative Arrangements

Who actually maintained the DNS and IP address roots? In the mid-1980s, as far as the available evidence indicates, this was not a question that generated much interest or controversy.

Postel's involvement in the definition of the new top-level domains made it clear that he and other researchers at ISI had been given by DARPA what would later be called the policy authority over name and number assignment. That is, they established the initial procedures for assigning and keeping track of protocol and network numbers, and decided what toplevel domains would be defined. A long-term contract between DARPA and ISI, which listed Postel as the principal investigator, contained a list of five or six work items, some of which were related to assignment functions but included other functions such as the RFC editor. ^{[24 ]}Postel's funding support from DARPA for those tasks would last until 1997.

The actual mechanics of registering domain names and addresses-the operational authority -was in different hands. Since 1971 the Stanford Research Institute (SRI) had maintained the hosts.txt file for the original ARPANET and the ARPA-Internet under the pre-DNS naming system. The services were performed under contract to the Defense Communications Agency and given the title Defense Data Network-Network Information Center (DDN-NIC). As domain-style names were introduced, the SRIoperated DDN-NIC retained its familiar role as the central point of coordination for the name space. It became the 'registrar of top-level and second-level domains, as well as administrator of the root domain name servers' for both the military and civilian parts of the Internet (RFC 1032, November 1987). ^{[25 ]}In November 1987, SRI's DDN-NIC also took over the IP address assignment and registry function from Postel and Reynolds at ISI (RFC 1020, November 1987). Both transfers of assignment authority followed a precedent in the Defense Department. Once a new system was no longer experimental, control was routinely transferred away from researchers to a military agency and put to practical use. The military agency might then contract with a private firm to perform various functions. ^{[26 ]}

In October 1982, the Defense Communications Agency decided to split the ARPA-Internet into two: the ARPANET would continue to connect academically based researchers supported by the military, while a separate, more restricted and secured MILNET would link military users. About half of the old ARPANET nodes went to MILNET. A few hosts were connected to both, as gateways for intercommunication. Name and number assignment functions for both networks, however, remained centralized at the DDN-NIC.

^[5]Cerf and Kahn, IEEE Transactions on Communication 22 (5): 637-648.

^{[6 ]}RFC 791, 'DARPA Internet Program Protocol Specification,' September 1981.

^{[7 ]}Joyce Reynolds, Jon Postel, RFC 870, 'Assigned Numbers,' October 1983.

^{[8 ]}Clark projected a future 'upper limit of about 1,000 networks.' RFC 814 (July 1982).

^{[9 ]}'[A] reasonable programming strategy would be to make the name table accessible only through a subroutine interface, rather than by scattering direct references to the table all through the code. In this way, it will be possible, at a later date, to replace the subroutine with one capable of making calls on remote name servers.' David Clark, RFC 814 (July 1982).

^{[10 ]}Archives of it still exist at <http://www.tcm.org/msggroup/>.

^{[11 ]}Jon Postel, 'Namedroppers Policy,' November 2, 1983, <http://ittf.vlsm.org/ietf/129.txt>.

^{[12 ]}'The general guideline for a second-level domain is that it have over 50 hosts. This is a very soft ‘requirement.' It makes sense that any major organization, such as a university or corporation, be allowed as a second-level domain-even if it has just a few hosts.' Postel, RFC 881 (November 1983).

^{[13 ]}'Top-level domains,' Postel wrote in RFC 881, 'must be specially authorized. In general, they will only be authorized for domains expected to have over 500 hosts.'

^{[14 ]}This May 11, 1984, draft of what became RFC 920 is available at <http:// ittf.vlsm.org/ietf/132.txt>.

^{[15 ]}Einar Stefferud to namedroppers list, May 13, 1984.

^{[16 ]}Mark Horton, namedroppers list, November 2, 1985.

^{[17 ]}Yet, Postel still faced criticism that the Internet administration was 'U.S.-centric' because some thought other countries had to use their country code as a top-level domain, whereas people in the United States didn't. Postel to namedroppers list, May 20, 1984, <http://ittf.vlsm.org/ietf/131.txt>. In arguing against this notion, Postel observed that anyone could register in .com, .edu, or .org.

^{[18 ]}Ibid.

^[19]An X.400 address has an eight-layer hierarchy, starting with a country code. Moreover, each X.400 messaging system is an independent domain and can only be interconnected by agreement among the implementing service providers. The features of the X.400 standard reflect its origins in a telephone monopoly- dominated world.

^{[20 ]}Jon Postel to msggroup, November 15, 1985.

^{[21 ]}Steve Kille to namedroppers list, November 15, 1985.

^{[22 ]}'Hi. The namedroppers list is for the discussion of the technical issues in the DARPA domain name system. The actual spelling of the name strings, and especially the semantics that people attach to those strings are not part of these technical issues. So please, no messages in this mailing list about the merits of EDU vs US (etc.) as a top-level domain name. Clearly, the choices of top-level names is a highly charged political issue. Please discuss it in the appropriate forum ( msggroup?, poli-sci??).-jon.'

^{[23 ]}Jon Postel, Namedroppers Policy, August 2, 1987, <http://ittf.vlsm.org/ietf/165.txt>.

^{[24 ]}'There was a contract executed in 1988 with DARPA, which I have seen. It contains a set of about 5 or 6 work items which are recognizable as the IANA functions plus the RFC editor and the Internet Monthly Report (which Postel also did). The term 'IANA' was not used in the contract. It's my recollection that this contract was preceded by another long-term contract between DARPA and ISI that included the same functions, but I've never seen that one.' Brian Kahin, email to author, December 19, 2000.

^{[25 ]}By maintaining both the DNS root and several different top-level domains (.arpa, .com, .edu, .org, .gov, and .mil) the DDN-NIC combined functions that, in a strict implementation of a hierarchical name space, should have been separate. Postel and other members of the technical community recognized this and didn't particularly like it, but accepted it because there was no one else to do the task. See Postel to namedroppers list, May 20, 1984, <http://ittf.vlsm.org/ietf/131.txt>.

^{[26 ]}As an example, responsibility for ARPANET management was transferred from ARPA to the Defense Communications Agency in 1975. See NAS (1994), 238, and Abbate (2000), 136.

---