ONLY CONNECT: Researcher [left] explains computer networking to French officials at a meeting in 1974. Zimmermann would later play a key role in the development of the Open Systems Interconnection standards.

08.OSIHistory.NA.indd 38 7/10/13 9:31 AM The that wasn’t The making—and forgetting—of the Open Systems Interconnection standards By Andrew L. Russell

If everything had gone according to plan, the Internet as we know it would never have sprung up. That plan, devised 35 years ago, instead would have created a comprehensive set of standards for computer networks called Open Systems Interconnection, or OSI. Its ar­ chitects were a dedicated group of computer industry representatives in the United King­ dom, France, and the United States who envisioned a complete, open, and multilayered­ system that would allow users all over the world to exchange data easily and thereby un­ leash new possibilities for collaboration and commerce. For a time, their vision seemed like the right one. Thousands of engineers and policy­ makers around the world became involved in the effort to establish OSI standards. They soon had the support of everyone who mattered: computer companies, telephone com­ panies, regulators, national governments, international standards setting agencies, aca­ demic researchers, even the U.S. Department of Defense. By the mid-1980s the worldwide adoption of OSI appeared inevitable. And yet, by the early 1990s, the project had all but stalled in the face of a cheap and ag­ ile, if less comprehensive, alternative: the Internet’s Transmission Control Protocol and Internet Protocol. As OSI faltered, one of the Internet’s chief advocates, Einar Stefferud, gleefully pronounced: “OSI is a beautiful dream, and TCP/IP is living it!” What happened to the “beautiful dream”? While the Internet’s triumphant story has been well documented by its designers and the historians they have worked with, OSI has been forgotten by all but a handful of veterans of the Internet-OSI standards wars. To

INRIA understand why, we need to dive into the early history of computer networking, a time

SPECTRUM.IEEE.ORG | north american | AUG 2013 | 39

08.OSIHistory.NA.indd 39 7/10/13 9:31 AM when the vexing problems of digital conver­ ment that some form of interna­ gence and global interconnection were very tional standardization would be [Baran] [davies] 1965: Donald W. Davies, much on the minds of computer scientists, necessary for to working independently telecom engineers, policymakers, and in­ be viable. An ­early attempt began in of Baran, conceives his dustry executives. And to appreciate that 1972, with the formation of the Inter­ “packet-switching” network. history, you’ll have to set aside for a few national Network Working Group minutes what you already know about the (INWG). was its first chair­ Internet. Try to imagine, if you can, that the man; other active members included 1961: Internet never existed. Alex ­McKenzie in the United States, at Rand Corp. begins to outline ­ and Roger Scantlebury­ his concept of The story starts in the 1960s. The Berlin in England, and Louis Pouzin and “message block Wall was going up. The Free Speech move­ ­Hubert Zimmermann in France. switching” as a way 1969: ARPANET, the first packet- of sending data over switching network, is created in ment was blossoming in Berkeley. U.S. The purpose of INWG was to pro­ computer networks. the United States. troops were fighting in Vietnam. And digi­ mote the “” style of packet tal computer-communication systems were switching that Pouzin had designed. in their infancy and the subject of intense, As he explained to me when we met wide-ranging investigations, with dozens in Paris in 2012, “The essence of data­ (and soon hundreds) of people in academia, gram is connectionless. That means 1970: Estimated U.S. market revenues for computer industry, and government pursuing major you have no relationship established communications: US $46 million. research programs. between sender and receiver. Things The most promising of these involved a just go separately, one by one, like

new approach to data communication called photons.” It was a radical proposal, [pouzin] [cerf] [mckenzie] [zimmermaNn] [davies] packet switching. Invented independently­ by especially when compared to the Paul Baran at the Rand Corp. in the United­ connection-oriented virtual cir­ States and Donald Davies at the National­ cuits favored by IBM and the tele­ Physical Laboratory in England, packet com engineers. switching broke messages into discrete INWG met regularly and ex­ blocks, or packets, that could be routed sep­ changed technical papers in an effort 1972: International Network Working Group (INWG) arately across a network’s various channels. to reconcile its designs for datagram forms to develop an international standard for packet- switching networks, with Vint Cerf as chairman. A computer at the receiving end would reas­ networks, in particular for a trans­ semble the packets into their original form. port protocol—the key mechanism for Baran and Davies both believed that packet exchanging packets across different switching could be more robust and efficient types of networks. After several years than circuit switching, the old technology of debate and discussion, the group finally drained from Cyclades between 1975 and used in telephone systems that required a reached an agreement in 1975, and Cerf and 1978, when Pouzin’s involvement also ended. dedicated channel for each conversation. Pouzin submitted their protocol to the inter­ For his part, Cerf was so discouraged by Researchers sponsored by the U.S. Depart­ national body responsible for overseeing tele­ his international adventures in standards ment of Defense’s Advanced Research Proj­ communication standards, the International making that he resigned his position as ects Agency created the first packet-switched Telegraph and Telephone Consultative Com­ INWG chair in late 1975. He also quit the network, called the ARPANET, in 1969. Soon mittee (known by its French acronym, CCITT). faculty at Stanford and accepted an offer other institutions, most notably the ­computer The committee, dominated by telecom to work with at ARPA. Cerf and giant IBM and several of the telephone mo­ engineers, rejected the INWG’s proposal Kahn had already drawn on Pouzin’s data­ nopolies in Europe, hatched their own am­ as too risky and untested. Cerf and his col­ gram design and published the details of bitious plans for packet-switched networks. leagues were bitterly disappointed. Pouzin, their “transmission control program” the Even as these institutions contemplated the the combative leader of Cyclades, France’s previous year in the IEEE Transactions on digital convergence of computing and com­ own packet-­switching research project, sar­ Communications. That provided the tech­ munications, however, they were anxious to castically noted that members of the CCITT nical foundation of the “Internet,” a term protect the revenues generated by their ex­ “do not object to packet switching, as long as adopted later to refer to a network of net­ isting businesses. As a result, IBM and the it looks just like circuit switching.” And when works that utilized ARPA’s TCP/IP. In sub­ telephone monopolies favored packet switch­ Pouzin complained at major conferences sequent years the two men directed the ing that relied on “virtual circuits”—a design about the “arm-twisting” tactics of “national development of Internet protocols in an that mimicked circuit switching’s technical monopolies,” everyone knew he was refer­ environment they could control: the small and organizational routines. ring to the French telecom authority. French community of ARPA contractors. With so many interested parties putting bureaucrats did not appreciate their country­ Cerf’s departure marked a rift within forth ideas, there was widespread agree­ man’s candor, and government funding was the INWG. While Cerf and other ARPA con­

40 | AUG 2013 | north american | SPECTRUM.IEEE.ORG

08.OSIHistory.NA.indd 40 7/10/13 9:31 AM A Brief History of the OSI Standards

[davies] January 1983: U.S. Department of Defense’s 2013: IPv6 carries 1971: Cyclades mandated use of TCP/IP on the ARPANET approximately 1 percent of packet-switching signals the “birth of the Internet.” global Internet traffic. project launches in [BACHMAN] [zimmermanN] [day] France. 1985: U.S. National Research Council May 1983: ISO recommends that the Department of Defense 1975: INWG publishes “ISO migrate gradually from TCP/IP to OSI. submits a proposal 7498: The Basic to the International Reference Model 1989: As OSI begins Telegraph for Open Systems to founder, computer www and Telephone 1977: International Organization for Standardization Interconnection” scientist Brian 1991: Tim Berners- Consultative (ISO) committee on Open Systems Interconnection is as an international Carpenter gives a talk Lee announces Committee (CCITT), formed, with Charles Bachman as chairman. standard. entitled “Is OSI Too public release of which rejects it. Cerf Late?” He receives a the WorldWideWeb resigns from INWG. standing ovation. application.

1988: U.S. market revenues IPv6 1974: IBM launches a packet-switching network 1980: U.S. Department for computer 1996: Internet called the Systems Network Architecture. of Defense publishes communications: community defines “Standards for the $4.9 billion. 1992: U.S. IP version 6. Internet Protocol and National Science Transmission Control Foundation revises Protocol.” policies to allow [cerf] [kahn] commercial traffic over the Internet. 1976: CCITT publishes 1988: U.S. Department Recommendation X.25, of Commerce mandates 1974: Cerf and Robert Kahn publish “A Protocol a standard for packet that government 1992: In a “palace revolt,” Internet engineers for Packet Network Intercommunication,” in IEEE switching that uses agencies buy OSI- reject the ISO ConnectionLess Network Transactions on Communications. “virtual circuits.” compliant products. Protocol as a replacement for IP version 4.

Baran: ; Davies: NPL; MAP and B­ achman: tractors eventually formed the core of the TC 1 for standards on screw threads and TC 17 Computer History Museum; Zimmermann and Day: John Day; Pouzin: Marc Weber/Computer History Museum; Cerf: Jose ­Internet community in the 1980s, many of for steel. Also unlike the CCITT, ISO already Mercado/Stanford News Service; McKenzie: Alex McKenzie; the remaining veterans of INWG regrouped had committees for computer standards Kahn: Louis F. Bachrach and joined the international alliance taking and seemed far more likely to be receptive shape under the banner of OSI. The two to connectionless . ­Bachman had four years earlier received camps became bitter rivals. The British proposal, which had the sup­ the prestigious Turing Award for his work port of U.S. and French representatives, on a database management system called OSI was devised by committee, but that called for “network standards needed for the Integrated Data Store. fact alone wasn’t enough to doom the open working.” These standards would, When I interviewed Bachman in 2011, he ­project—after all, plenty of successful stan­ the British argued, provide an alternative described the “architectural vision” that he dards start out that way. Still, it is worth not­ to traditional computing’s “self-contained, brought to OSI, a vision that was inspired ing for what came later. ‘closed’ systems,” which were designed with by his work with databases generally and In 1977, representatives from the British “little regard for the possibility of their inter­ by IBM’s Systems Network Architecture in computer industry proposed the creation working with each other.” The concept of particular. He began by specifying a refer­ of a new standards committee devoted to open working was as much strategic as it ence model that divided the various tasks of packet-switching networks within the Inter­ was technical, signaling their desire to en­ computer communication into distinct layers. national Organization for Standardization able competition with the big incumbents— For example, physical media (such as copper (ISO), an independent nongovernmental namely, IBM and the telecom monopolies. cables) fit into layer 1; transport protocols for ­association created after World War II. Unlike As expected, ISO approved the British re­ moving data fit into layer 4; and applications the CCITT, ISO wasn’t specifically concerned quest and named the U.S. database expert­ (such as e-mail and file transfer) fit into layer 7. with telecommunications—the wide-ranging Charles Bachman as committee chairman. Once a layered architecture was established, topics of its technical committees included Widely respected in computer circles, specific protocols would then be developed.

SPECTRUM.IEEE.ORG | north american | AUG 2013 | 41

08.OSIHistory.NA.indd 41 7/10/13 9:31 AM OSI vs. TCP/IP ed had market interests to worldviews. Zimmermann and his comput­ Application protect and pet projects to ing colleagues, inspired by Pouzin’s data­ presentation Application advance. Delegates from the gram design, championed “connectionless”

session same country often had di­ protocols, while the telecom professionals vergent agendas. Many at­ persisted with their virtual circuits. Instead transport transport tendees were veterans of of resolving the dispute, they agreed to in­ network internet INWG who retained a wary clude options for both designs within OSI, data link optimism that the future of thus increasing its size and complexity. Network access physical data networking could be This uneasy alliance of computer and wrested from the hands of telecom engineers published the OSI refer­

A layered approach: The OSI reference model [left column] di- IBM and the telecom mo­ ence model as an international standard in vides computer communications into seven distinct layers, from phys- nopolies, which had clear 1984. Individual OSI standards for transport ical media in layer 1 to applications in layer 7. Though less rigid, the intentions of dominating protocols, electronic mail, electronic direc­ TCP/IP approach to networking can also be construed in layers, as shown on the right. this emerging market. tories, network management, and many Meanwhile, IBM represen­ other functions soon followed. OSI began Bachman’s design departed from IBM’s tatives, led by the company’s capable direc­ to accumulate the trappings of inevitabil­ Systems Network Architecture in a signifi­ tor of standards, Joseph De Blasi, masterfully ity. Leading computer companies such as cant way: Where IBM specified a terminal- steered the discussion, keeping OSI’s devel­ Digital Equipment Corp., Honeywell, and to-­computer architecture, Bachman would opment in line with IBM’s own business in­ IBM were by then heavily invested in OSI, connect computers to one another, as peers. terests. Computer scientist John Day, who as was the European Economic Community That made it extremely attractive to compa­ designed protocols for the ARPANET, was and national governments throughout Eu­ nies like General Motors, a leading proponent a key member of the U.S. delegation. In his rope, North America, and Asia. of OSI in the 1980s. GM had dozens of plants 2008 book Patterns in Network Architecture Even the U.S. government—the main spon­ and hundreds of suppliers, using a mix of (Prentice Hall), Day recalled that IBM repre­ sor of the Internet protocols, which were largely incompatible hardware and software. sentatives expertly intervened in disputes incompatible with OSI—jumped on the OSI Bachman’s scheme would allow “interwork­ between delegates “fighting over who would bandwagon. The Defense Department offi­ ing” between different types of proprietary get a piece of the pie.… IBM played them cially embraced the conclusions of a 1985 computers and networks—so long as they like a violin. It was truly magical to watch.” National Research Council recommenda­ followed OSI’s standard protocols. Despite such stalling tactics, Bachman’s tion to transition away from TCP/IP and to­ The layered OSI reference model also pro­ leadership propelled OSI along the precari­ ward OSI. Meanwhile, the Department of vided an important organizational feature: ous path from vision to reality. ­Bachman and Commerce issued a mandate in 1988 that modularity. That is, the layering allowed Hubert Zimmermann (a veteran of Cyclades­ the OSI standard be used in all computers committees to subdivide the work. Indeed, and INWG) forged an alliance with the tele­ purchased by U.S. government agencies Bachman’s reference model was just a start­ com engineers in CCITT. But the partner­ ­after August 1990. ing point. To become an international stan­ ship struggled to overcome the fundamental While such edicts may sound like the work dard, each proposal would have to complete incompatibility between their respective of overreaching bureaucrats, remember that a four-step process, starting throughout the 1980s, the with a working draft, then a ­Internet was still a research net­ draft proposed international work: It was growing rapidly, to standard, then a draft inter­ be sure, but its managers did national standard, and final­ not allow commercial traffic ly an international standard. or for-profit service providers Building consensus around the on the government­ -subsidized OSI reference model and asso­ backbone until 1992. For busi­ ciated standards required an nesses and other large enti­ extra­ordinary number of ple­ ties that wanted to exchange nary and committee meetings. data between different kinds OSI’s first plenary meet­ of computers or different types ing lasted three days, from of networks, OSI was the only 28 February through 2 March game in town. 1978. Dozens of delegates from 10 countries participat­ That was not the end of the sto­ ed, as well as observers from What’s In A Name: At a July 1986 meeting in Newport, R.I., representatives ry, of course. By the late 1980s, from France, Germany, the United Kingdom, and the United States consid- four international organiza­ ered how the OSI reference model would handle the crucial functions of nam- frustration with OSI’s slow de­ D a y

tions. Everyone who attend­ ing and addressing on the network. velopment had reached a boil­ Joh n

42 | AUG 2013 | north american | SPECTRUM.IEEE.ORG

08.OSIHistory.NA.indd 42 7/10/13 9:31 AM ing point. At a 1989 meeting in Europe, the odds. And so openness and modularity—the ment subsidies for their research eventually OSI advocate Brian Carpenter gave a talk ti­ key principles for coordinating­ the project— created a distinct commercial advantage: tled “Is OSI Too Late?” It was, he recalled in ended up killing OSI. Internet protocols could be implemented a recent memoir, “the only time in my life” for free. (To use OSI standards, companies that he “got a standing ovation in a techni­ Meanwhile, the Internet flourished. With that made and sold networking equipment cal conference.” Two years later, the French ample funding from the U.S. government, had to purchase paper copies from the stan­ networking expert and former INWG mem­ Cerf, Kahn, and their colleagues were shield­ dards group ISO, one copy at a time.) Marc ber Pouzin, in an essay titled “Ten Years of ed from the forces of international politics Levilion, an engineer for IBM France, told OSI—Maturity or Infancy?,” summed up the and economics. ARPA and the Defense Com­ me in a 2012 interview about the computer growing uncertainty: “Government and cor­ munications Agency accelerated the Inter­ industry’s shift away from OSI and toward porate policies never fail to recommend OSI net’s adoption in the early 1980s, when they TCP/IP: “On one side you have something as the solution. But, it is easier and quicker to subsidized researchers to implement Inter­ that’s free, available, you just have to load it. implement homogenous networks based on net protocols in popular operating systems, And on the other side, you have something proprietary architectures, or else to intercon­ such as the modification of Unix by the Uni­ which is much more architectured, much nect heterogeneous systems with TCP-based versity of California, Berkeley. Then, on more complete, much more elaborate, but products.” Even for OSI’s champions, the 1 January 1983, ARPA stopped supporting it is expensive. If you are a director of com­ Internet was looking increasingly attractive. the ARPANET­ host protocol, thus forcing its putation in a company, what do you choose?” That sense of doom deepened, progress contractors to adopt TCP/IP if they wanted By the mid-1990s, the Internet had become stalled, and in the mid-1990s, OSI’s beauti­ to stay connected; that date became known the de facto standard for global computer ful dream finally ended. The effort’s fatal as the “birth of the Internet.” networking. Cruelly for OSI’s creators, In­ flaw, ironically, grew from its commitment And so, while many users still expected ternet advocates seized the mantle of “open­ to openness. The formal rules for interna­ OSI to become the future solution to global ness” and claimed it as their own. Today, they tional standardization gave any interested network interconnection, growing numbers routinely campaign to preserve the “open party the right to participate in the design began using TCP/IP to meet the practical Internet” from authoritarian governments, process, thereby inviting structural tensions, near-term pressures for interoperability. regulators, and would-be monopolists. incompatible visions, and disruptive tactics. Engineers who joined the Internet com­ OSI’s first chairman, Bachman, had an­ munity in the 1980s frequently misconstrued In light of the success of the nimble Internet, ticipated such problems from the start. In OSI, lampooning it as a misguided monstros­ OSI is often portrayed as a cautionary tale of a conference talk in 1978, he worried about ity created by clueless European bureaucrats. overbureaucratized “anticipatory standard­ OSI’s chances of success: “The organization­ Internet engineer Marshall Rose wrote in ization” in an immature and volatile market. al problem alone is incredible. The technical his 1990 textbook that the “Internet com­ This emphasis on its failings, however, ­misses problem is bigger than any one previously munity tries its very best to ignore the OSI OSI’s many successes: It focused attention on faced in information systems. And the po­ community. By and large, OSI technology is cutting-edge technological questions, and litical problems will challenge the most as­ ugly in comparison to Internet technology.” it became a source of learning by doing—­ tute statesmen. Can you imagine trying to Unfortunately, the Internet community’s including some hard knocks—for a genera­ get the representatives from ten major and bias also led it to reject any technical insights tion of network engineers, who went on to competing computer corporations, and ten from OSI. The classic example was the “pal­ create new companies, advise governments, telephone companies and PTTs [state-owned ace revolt” of 1992. Though not nearly as and teach in universities around the world. telecom monopolies], and the technical ex­ formal as the bureaucracy that devised OSI, Beyond these simplistic declarations of perts from ten different nations to come to the Internet had its Internet Activities Board “success” and “failure,” OSI’s history holds any agreement within the foreseeable future?” and the Internet Engineering Task Force, important lessons that engineers, policymak­ Despite Bachman’s and others’ best ef­ responsible for shepherding the develop­ ers, and Internet users should get to know forts, the burden of organizational overhead ment of its standards. Such work went on better. Perhaps the most important lesson never lifted. Hundreds of engineers ­attended at a July 1992 meeting in Cambridge, Mass. is that “openness” is full of contradictions. the meetings of OSI’s various committees Several leaders, pressed to revise routing OSI brought to light the deep incompatibility and working groups, and the bureaucratic and ­addressing limitations that had not between idealistic visions of openness and procedures used to structure the discussions been anticipated when TCP and IP were the political and economic realities of the didn’t allow for the speedy production of designed, recommended that the commu­ international networking industry. And OSI standards. Everything was up for debate— nity ­consider—if not adopt—some technical eventually collapsed because it could not rec­ even trivial nuances of language, like the protocols developed within OSI. The hun­ oncile the divergent desires of all the interest­ difference between “you will comply” and dreds of Internet engineers in attendance ed parties. What then does this mean for the “you should comply,” triggered complaints. howled in protest and then sacked their continued viability of the open Internet? n More significant rifts continued between leaders for their heresy.

OSI’s computer and telecom experts, whose Although Cerf and Kahn did not design Post your comments online at technical and business plans remained at TCP/IP for business use, decades of govern­ http://spectrum.ieee.org/osihistory0813­

SPECTRUM.IEEE.ORG | north american | AUG 2013 | 43

08.OSIHistory.NA.indd 43 7/10/13 9:31 AM