In the early 1970s, working as a researcher for the French government, Pouzin created a computer network known as CYCLADES, and Vint Cerf himself has cited Pouzin’s design as one of the key influences behind the development of TCP/IP.
It isn’t hard to see why. With CYCLADES, Pouzin built a network where the delivery of information between machines was overseen by the machines themselves — not by some piece of central network hardware. In other words, he realized one the fundamental ideas that makes the Internet the Internet.
“We designed CYCLADES to be connected to other networks — in the future,” Pouzin remembers.
This past April, in recognition of his role in the creation of TCP/IP and his contribution to various other networking standards, Pouzin was inducted into the Internet Society’s (ISOC) Internet Hall of Fame. Part of the Hall’s inaugural class, he was enshrined alongside such as names as Sir Tim Berners-Lee, Ray Tomlinson, Leonard Kleinrock, and, yes, Vint Cerf and Bob Kahn.
Louis Pouzin was born in France and studied at the École Polytechnique in Paris. But before designing CYCLADES on behalf of the Delegation a l’Informatique — the French government agency that oversaw computing projects — he played a part in the development of a seminal computing system here in the United States. At the Massachusetts Institute of Technology in the mid-1960s, he worked on the Compatible Time-Sharing System, or CTSS, one of the world’s first time-sharing systems. Time-sharing was the forerunner of modern networking, letting multiple people tap into a single machine using remote terminals.
As a member of the computer center staff at MIT, Pouzin wrote dozens of CTSS commands from scratch, before realizing it made more sense to build a program that would help automate that creation of such commands. The result was RUNCOM, the first operating system “shell” — i.e. a higher-level software interface that provides access to the core of the OS. It was Pouzin who coined the term “shell,” and RUNCOM that would eventually give rise to the first shell that ran atop UNIX, the operating system that would spread the idea across the computing world. But Pouzin would have a larger effect on the rise of computer networking.
In joining the Delegation a l’Informatique at the end of 1971, Pouzin was charged with building a national research network, and he promptly returned to the U.S. to meet with several of the minds behind the ARPANET — the research network the U.S. Department of Defense has launch at the tail end of 60s. This included Larry Roberts, who originally designed the network; Cerf, who was part of the team that setup the first node at the University of California, Los Angeles; and Leonard Kleinrock, who had led that team — and who Pouzin had known during his days at MIT.
In building CYCLADES, Pouzin used the ARPANET as a model, as well as the British research network overseen by Donald Davies across the English Channel. Like the ARPANET and Davies’ National Physical Laboratory network, CYCLADES used packet-switching, meaning information would be broken down into tiny messages, or packets, before traveling across the wire. But Pouzin reinvented these network packets, creating something called the “datagram.”
In a departure from earlier packet-switching networks, a datagram contained all the information needed to route it between two machines — i.e. it didn’t require any information related to previous exchanges between the two machines. There was no confirmation of delivery or, indeed, no guarantee of delivery. But datagrams could arrive at their destinations in any order and at any time — and they gave client machines the power to delivery data without oversight by some central authority.
Donald Davies had run simulations that used something akin to datagrams, but with CYCLADES, Pouzin was the first to make them a reality. Similar datagrams were then adopted by Cerf and Kahn when they designed TCP/IP in the mid-1970s. Pouzin, Cerf, and Kahn were all part of an international working group that was then striving to build networks that could freely exchange information.
Cerf compares the datagram setup to the delivery of old school postcards through the U.S. Mail. “There’s a ‘from’ address and a ‘to’ address and some content, and you send them and — somehow — they get to the other end,” he says. “But there’s no guarantee they get to the other end. It’s a best-effort service. And if you send two messages, they may not emerge at the other end in the same order. Basically, it’s a very fast but not necessarily guaranteed or orderly service.”
On the face of it, that may not seem like a particularly attractive proposition. But the key is that it’s simple — and that it lets you build a network that can span all sorts of other networks. Once this basic arrangement is in place, you can always layer on additional tools that provide guarantees and order, and whether else you may want. You can layer TCP atop IP. And you can build an Internet.
That’s what happened. And we have Louis Pouzin to thank for it.
