‘Rough Consensus and Running Code’ and the Internet-OSI Standards War Andrew L. Russell The Johns Hopkins University Internet historians recognize the technical achievements but often overlook the bureaucratic innovations of Internet pioneers. The phrase, “We reject: kings, presidents, and voting. We believe in: rough consensus and running code,” was coined by David Clark in 1992. This article explains how the phrase captured the technical and political values of Internet engineers during a crucial phase in the Internet’s growth. Historians interested in conflict and consensus overlooked, however, is the extent to which the in technological systems stand much to gain success of the Internet depended on organiza- from examining standardization—the process- tional innovations directed by Cerf, Kahn, and es of creating and implementing technical stan- other Internet pioneers such as Jon Postel and dards. Standardization requires sophisticated David Clark. Starting in the 1970s, researchers technical work as well as organizational and in groups such as the Internet Configuration strategic coordination. In the late 1970s, it was Control Board (ICCB), the Internet Activities reasonable to assume that formal standards-set- Board (IAB), and the Internet Engineering Task ting institutions, such as the International Force (IETF) coordinated Internet standards Telecommunication Union (ITU) and the and architectural development. These institu- International Organization for Standardization tions deserve special attention not only for (ISO), would lead the coordination and stan- their technical achievements, but also because dardization of information services, much as they fostered a voluntary international con- they did for international telecommunications. sensus during a period of intense technical Beginning in 1977, ISO oversaw a large, ambi- and institutional change in computing and tious attempt to define a network architecture telecommunications. called Open Systems Interconnection (OSI). These organizational innovations emerged Between the late 1970s and the early 1990s, the as responses to external competition from ISO OSI seven-layer model became enshrined in as well as to internal strains in the processes of computer science curricula and endorsed by Internet standardization. As the Internet grew governments around the world. Competing rapidly in the late 1980s and 1990s, groups such networks—including experimental TCP/IP net- as the IAB and IETF started to become victims works—were expected to fade away once OSI of their own success and struggled to preserve protocols were standardized and implemented their founding principles. If we understand by users and manufacturers. “politics” to mean relations of control between By the mid-1990s, however, ISO’s slow stan- individuals and groups, the political values of dardization process had failed to keep up with Internet architects and engineers were especial- alternative, informal mechanisms that were ly evident as these people built political struc- more effective at coordinating rapid techno- tures—standards bodies—from scratch.3 logical change.1 These informal mechanisms— Standardization is technically and organiza- the focus of this article—provided vital tionally complex as well as deeply value-laden.4 institutional support for the eventual success Leaders and participants in the IAB and IETF and global deployment of the Internet archi- articulated institutional rules, cultural tradi- tecture and Internet standards such as TCP/IP. tions, and versions of their own history as they In most existing histories of the Internet, responded to challenges from within and with- leaders such as Vinton Cerf and Robert Kahn out. A memorable phrase encapsulates the pre- receive plaudits for their technical work, espe- vailing technical and organizational values of cially for their roles in the creation of the core those who were involved in the Internet stan- Internet standards TCP and IP.2 What is often dards process from the mid-1970s to the pres- 48 IEEE Annals of the History of Computing Published by the IEEE Computer Society 1058-6180/06/$20.00 © 2006 IEEE ent: “We reject: kings, presidents, and voting. Arpanet (PRnet was developed in Hawaii), but We believe in: rough consensus and running faced a problem of trying to connect what he code.”5 This article explains how this phrase— later called “two radically different networks” which became a motto for Internet standardi- with different network capacities, protocols, zation—articulated a common set of beliefs and routing systems.10 ARPA’s concurrent about the work culture and engineering style of development of satellite packet switching in Internet standardization. SATnet further compounded this problem, In 1999, law professor Lawrence Lessig leading Kahn to conclude that network inter- declared that “rough consensus and running connection could not be achieved on an ad code” had broad significance as “a manifesto hoc, network-to-network level. that will define our generation.”6 An examina- To overcome this problem, Kahn in 1973 tion of the origins of this credo—coined by rekindled an effective partnership with Vint David Clark in 1992—illustrates some techni- Cerf—the two had worked together on the first cal, rhetorical, and philosophical differences nodes of the Arpanet in 1969—and proposed a between the Internet standards process and the new way to transport data packets. This mech- competing ISO process. Most participants in the anism was a simple technical protocol (trans- Internet standards process today consider mission control protocol, or TCP) and system “rough consensus and running code” to be a of gateways (now known as routers) to transfer succinct and accurate description of the data packets between the dissimilar networks.11 Internet standards process.7 For historians, Kahn’s ultimate goal was to make the network “rough consensus and running code” stands as transparent, “a seamless whole,” invisible to a revealing depiction of the international poli- the user who would be more interested in tics of computer internetworking in the last accessing information over the network instead quarter of the 20th century, as well as a point of of the operation of the network itself.12 This entry for investigating why contemporaries basic principle—interconnection via standard described the competition between the Internet protocols—is the Internet’s fundamental struc- and OSI as “the Internet religious war.”8 ture and defining feature.13 In 1978, Cerf and two other DARPA-funded network researchers, Internet architecture: TCP/IP and the Danny Cohen and Jon Postel, split the func- end-to-end argument tions of TCP into two protocols, TCP and the In 1996, the Internet Architecture Board Internet Protocol (IP), that worked together in (IAB) published RFC 1958, “Architectural the now-familiar combination of TCP/IP.14 Principles of the Internet,” to record aspects of Network transparency and application architectural approach practiced in the Internet autonomy are the two key elements of a design community over the previous 25 years. They philosophy first articulated in 1981 by David wrote: “in very general terms, the community Clark, Jerome Saltzer, and David Reed, three believes that the goal is connectivity, the tool veterans of ARPA-sponsored networking exper- is the Internet Protocol, and the intelligence is iments at the Massachusetts Institute of end to end rather than hidden in the net- Technology’s (MIT’s) Laboratory for Computer work.”9 This section provides a brief overview Science. Their paper, “End-To-End Arguments of the distinctive aspects of Internet architec- in System Design,” was written to explore a ture by exploring these three principles: inter- design principle that they claimed had “been connection, the Internet Protocol, and the used for many years with[out] explicit recog- “end-to-end” argument. nition.”15 An outgrowth of substantial experi- After initial networking experiments with ence with TCP/IP networking, the end-to-end the Arpanet beginning in 1969, the Advanced principle held that the Internet’s complex Projects Research Agency (ARPA) continued to functions should be performed at the end- explore networking concepts for satellite and points, leaving only the (relatively) simple radio. The problem of how to enable commu- tasks of interconnection and data transport to nication between technically disparate systems the network. The authors commended the also had implications for mobile military com- technical merits and simplicity of the end-to- munications. In the early 1970s, ARPA devel- end model by concluding, “end-to-end argu- oped a packet radio network (known as PRnet) ments are a kind of ‘Occam’s razor’ when it based on the same packet-switching technolo- comes to choosing the functions to be provid- gy being tested in the Arpanet. Robert Kahn, as ed in a communications subsystem.”16 program manager for this project, identified The end-to-end design principle thus calls the need to connect packet radio networks to for a simple standard for data transfer that large computers in the continental US via the allows new innovations to be added at the July–September 2006 49 ‘Rough Consensus and Running Code’ edges and on top of the “stupid network.”17 sion formally to expand the involvement of the From this distributed design follows decentral- research community in making decisions about ized control, both over the functionality of the the network: network and over the content of the network traffic. Marjory