Skip to content

Hubert Zimmermann and the Seven Layers

Abstract

Every networking course still opens with the same diagram: physical, data link, network, transport, session, presentation, application. The seven layers come from an ISO committee that met from 1978, and the man who wrote them up and argued them through the meetings was Hubert Zimmermann (1941–2012), a French engineer who had learned packet switching in Louis Pouzin’s CYCLADES group. He spent a decade building a set of standards that almost nobody deployed, and produced in passing the vocabulary that everyone, including the people who beat him, uses to this day.

Rocquencourt

Zimmermann was born on 15 November 1941 and studied at the École Polytechnique and the École Nationale Supérieure des Télécommunications. In 1972 he joined IRIA at Rocquencourt, west of Paris, and went to work for Louis Pouzin on CYCLADES, the research network that carried unreliable datagrams in its CIGALE subnet and pushed all responsibility for reliability out to the host computers.

He was in the International Network Working Group from its founding in 1972, the loose committee where the ARPANET, CYCLADES and NPL people compared protocol designs. That is where his name first travelled: when Vint Cerf and Bob Kahn published “A Protocol for Packet Network Intercommunication” in 1974, the paper that laid out TCP, their acknowledgements thanked “R. Metcalfe, R. Scantlebury, D. Walden, and H. Zimmerman” for comments on international network protocols, with his name misspelled. See Vint Cerf and Bob Kahn.

The Committee

In 1977 British delegates proposed a new ISO committee for the standards “needed for open working”, with American and French support. ISO agreed and made Charles Bachman, the database designer, its chairman. The first plenary ran from 28 February to 2 March 1978 with delegates from ten countries.

The problem the committee was set was real. IBM’s SNA connected terminals to an IBM mainframe, DEC’s DECnet connected DEC machines, and a company like General Motors that had bought from several vendors could not get its factories to talk to each other. The answer Bachman sketched was a stack of layers, each with a defined job, from the physical medium at layer 1 to applications such as mail and file transfer at layer 7. The design let the committee split itself along the same lines and work on the layers in parallel.

Zimmermann became the model’s chief author and its public advocate. Bachman, who chaired the committee, later called him “one of the very most important people on that committee, maybe the most important person, in terms of the contributions to it”. He argued for connectionless service, the datagram idea he had brought from Rocquencourt, against the telephone engineers of the CCITT, who wanted virtual circuits and had already rejected the INWG’s datagram proposal in 1975 as too risky and untested. Pouzin’s summary of that meeting was that the delegates “do not object to packet switching, as long as it looks just like circuit switching”.

Bachman and Zimmermann settled the fight by not settling it. The reference model was written to accommodate both connectionless and connection-oriented operation at several layers, which kept the computer and telecom camps in the same room and doubled the number of things any implementer had to build.

The Paper

Zimmermann set out the architecture in “OSI Reference Model: The ISO Model of Architecture for Open Systems Interconnection”, published in IEEE Transactions on Communications in April 1980, four years before the model became an international standard as ISO 7498 in 1984. A second exposition with John Day followed in the Proceedings of the IEEE in December 1983. Those two papers are the reason the seven layers outlived the protocols: they explain the model as a way of thinking about networks rather than as a specification to be implemented, and that is how the model got into the textbooks.

The vocabulary won completely. Engineers who have never touched an OSI protocol say “layer 2 switch”, “layer 3 routing”, “layer 7 firewall” and “layer 8 problem” when they mean the user. TCP/IP has four layers of its own and is routinely described using the ISO numbering anyway.

Dead End: The Standards Nobody Ran

The protocols standardised under the model, X.400 for mail, X.500 for directories, FTAM for file transfer, arrived late, cost money to read, and were hard to implement. National governments tried to fix that by decree: a National Research Council panel recommended in 1985 that the US Defense Department move off TCP/IP and onto OSI, and GOSIP, issued as FIPS 146 in February 1989, made OSI compliance a condition of federal network purchases from August 1990. Agencies bought compliant equipment, kept running TCP/IP, and in May 1995 FIPS 146-2 quietly reopened the door by accepting other open standards, the IETF’s among them.

Andrew Russell, who wrote the standard history of the episode, puts the blame on the openness of the process itself. Formal international standardisation let everybody in, and everybody came: hundreds of engineers, ten competing computer firms and ten state telephone monopolies, working under rules that guaranteed a hearing for every objection and made speed impossible. Meanwhile Marc Levilion, an engineer at IBM France, could describe the competition in eight words: TCP/IP was “free, available, you just have to load it”. At a European meeting in 1989, Brian Carpenter, an OSI advocate himself, gave a talk called “Is OSI Too Late?” and got a standing ovation, the only one he ever had at a technical conference. See Early Networking Failures.

Zimmermann had left for France Télécom in 1980 and stayed until 1986, so he watched the collapse from outside the committee he had built.

Chorus

The second half of his career had nothing to do with networks on the wire. Zimmermann started the Chorus distributed operating system as a research project at INRIA in 1979, built around a small kernel that passed messages between ports while the rest of the system ran as ordinary processes. Three research versions later, in 1986, he founded Chorus Systèmes to sell it. Chorus and Mach were the two earliest microkernels, and Chorus found its commercial market in telecom switching equipment. Sun Microsystems bought the company in 1997 and Zimmermann spent five years there as director of telecom software engineering. See Dead End: Sun Microsystems and Operating System Concepts.

After Sun he financed and advised small French technology companies: Arbor Venture Management, UDcast, Gingko Networks, Boost Your StartUp. ACM SIGCOMM had given him its award in 1991 for “20 years of leadership in the development of computer networking and the advancement of international standardization”. He died on 9 November 2012, six days short of his seventy-first birthday.

📚 Sources