Friday morning, I took the first plane down to Nice to meet Christian Huitema, the first European member of the Internet Architecture Board. One parochial American had referred to him as the first “alien” on the IAB, but in my view that title had long since been taken by one of the American members.
In Nice, I picked up my rental car and joined the morning grand prix doing at least double all posted speed limits. In Antibes, I took the turnoff for Sophia-Antipolis, the “French Silicon Valley.” Sophia-Antipolis was founded 20 years ago as a joint venture between nine local towns, a regional government, and the Nice chamber of commerce. The regional group, the Departement des Alpes-Maritimes, contributes over half the money.
By 1991, the park had grown to over 700 firms, the majority in computers and communications. Air France keeps its central reservation system here, Digital keeps a major research center, and the EEC has the European Telecommunications Standards Institute (ETSI).
Sophia-Antipolis is also the site of one of the major centers of the Institut National de la Recherche en Informatique et Automatique (INRIA), the premier French computer science research institute and the home base of Christian Huitema.
Traveling too fast, I zoomed out of a roundabout and didn’t see the INRIA sign. A few kilometers later, I saw a sign for INRA and turned in. The place was full of greenhouses. I had arrived at the Institut National de la Recherche Agronomique. Woops.
After a few more diversions, I found INRIA. INRIA, like ORSTOM, is a national research institution. It specializes in computer science and numerical methods and employs 900 staff in 5 locations. Each location is organized into a series of research projects.
At the Sophia-Antipolis site, there were a couple of dozen projects employing a little over 300 staff members. Three projects were in robotics, including work for the European Space Agency’s Mars mission. Occasionally, the robot would take a tour around the building to test its navigation skills.
Christian headed a project in computer networks. When I finally made it into his office, he was explaining his work to a young German who was considering starting a Ph.D. at a nearby university and working at INRIA to do his research.
I spent the day with Christian, and by the end of the day I began to see why he had been named to the IAB. At the age of 22, Christian had no definite career plans, so he became a consultant. He did quite well, but found he was always doing the same thing and got bored.
He then joined the Centre National d’Etudes de Telecommunications (CNET). By 1985, he had received the Doctorate d’Etat, roughly equivalent to an assistant professor at a U.S. university. His thesis, part of the NADIR project, dealt with transport protocols for satellite links.
He had by then moved on to GIPSI, a French effort to build a workstation. CNET made the hardware, there was a port of UNIX, and the system even sported a graphical interface.
As part of the team working on GIPSI, Christian concentrated on networking. He invented an NFS-like remote disk protocol that worked directly on Ethernet. He also designed a version of X.25 that ran over Ethernet.
This modified X.25 was implemented by removing HDLC, since 802.3 already provided a satisfactory data link. Any transmission necessary was moved up into the link layer of X.25. The software even did call setup using an ARP-like mechanism. The calling station broadcast a call setup and the target station would respond.
The result was an X.25 running at over 500 kbps, matching the performance of TCP on the test platform, a Motorola 68010. Not bad, to say the least. Christian, backed by the French PTT, brought this idea to the international standards table. This was his first, but certainly not his last, encounter with what he sardonically calls the “classical” standards process.
The opposition camp, led by the English and the Germans, felt that if you did X.25, you should do so by the book. This meant that sending datagrams into the ether was out. Instead, they wanted the use of Logical Link Control class II, a connection-oriented version of Ethernet which has been widely standardized but rarely used.
The whole thing degenerated into a religious stalemate. As with many talented people, this encounter with the standards world moved Christian more firmly into the TCP/IP camp.
In 1986, Christian was brought on as an INRIA staff member. In an unusual move, INRIA immediately named him a project leader. One of the first projects in Christian’s group was an X.400 over X.25 implementation. Christian’s work on X.400 led to other projects such as an X.400/SMTP gateway. That gateway is still up and running at INRIA and is used by several groups. SWITCH, the Swiss Research Network, used a leased line to INRIA to link their X.400 services to the Internet SMTP base for several months. Over time, that mail link migrated into a 64 kbps IP line from SWITCH to INRIA.
Related to the X.400 project was an X.500 project called The Obviously Required Name Server (THORN), run under the auspices of the European Community Esprit project. THORN led to a DSA implementation called Pizarro. Naming it after the Spanish conqueror was a takeoff on the UCL’s Quipu, an Incan name.
Working on all these OSI-based systems gave Christian’s group considerable experience in coding the OSI middle layers. One outgrowth of that experience was an ASN.1 compiler.
ASN.1 is a very powerful, very general specification for the presentation layer in OSI. The problem is that most people write programs in C, not ASN.1.
When an application receives data from an OSI network, as in the case of an incoming X.400 message, that message is encoded in ASN.1. An ASN.1 compiler starts with the ASN.1 specification for a message and generates the encoding and decoding routines that allow an application to interact with the network.
The INRIA compiler, known as MAVROS, was a fully general system, accepting any ASN.1 specification. It also handled variants of ASN.1, such as the X.509 standard used for digital signatures. It supported a lightweight version of the presentation layer, quite useful when similar machines send lots of integers and floating point numbers back and forth.
An outgrowth of the compiler effort was an ASN.1 benchmark. To test the compiler, and to compare ASN.1 to other presentation paradigms such as the Sun XDR specification, the benchmark provides a basis for measuring performance.
Christian’s group started by examining a typical X.400 message. Based on that message, a tree was developed with a depth of 8 and a total of 400 ASN.1 elements. As per the typical X.400 message, 80 percent of the elements were octet strings, 10 percent strings, and 10 percent integers.
Next, they started comparing the code generated by the ASN.1 compiler with basic lightweight methods. They looked at the amount of time to encode and decode data as well as the amount of data transferred on the network.
Needless to say, the initial performance was slow. The results from the benchmark were used to start optimizing the compiler. For example, decoding is slow in ASN.1 because of the tag, length, and value encoding for each element. In a technique similar to Van Jacobson’s TCP work, the ASN.1 routines began using header prediction techniques. If the next tag can be predicted from the current context, header prediction can greatly speed up decoding.
Running the ISO layers over TCP/IP and FDDI on two DEC 5000s, the INRIA group was able to achieve presentation layer throughput of 8 Mbps. By contrast, TCP throughput on the same configuration was 17 Mbps, UDP 25 Mbps. The bottleneck in all three cases appeared to be the CPU.
In addition to his research efforts, Christian has been active in helping to link the French infrastructure to the Internet. As with many international connections, this one can be traced back to the ubiquitous Larry Landweber. Christian was invited to attend one of Landweber’s networking conferences, this one hosted by Dennis Jennings in Dublin in December 1986.
Larry and Christian started talking and decided that INRIA should be linked to the Internet. Larry went to NSF, Christian to INRLA and each got half the link funded. Before the link was up, UUCP mail to the U.S. could take anywhere from hours to days. With an initial link speed of 56 kbps, mail started taking just minutes, or less. The INRIA link to the U.S. gave Christian more involvement with the Internet community and led to his appointment to the IAB.
I had a wonderful meal of foie gras and veal, hoping there were no animal rights activists lurking outside. To drink, we had a nice Bordeaux followed by an even better Beaujolais. One thing I’ve noticed is that one of the qualifications for membership on the IAB seems to be a exquisite taste in wines (many of which come with exquisite prices). IAB members such as Stephen Kent, Vinton Cerf, Lyman Chapin, Dan Lynch, and Christian Huitema are all renowned in the Internet community for their technical abilities in this area.
Saturday morning, I retrieved my car and drove along the coast of the Mediterranean back to the Nice airport. Arriving back at Hertz 26 hours after I had picked up my keys, I was handed an invoice for U.S. $300.
“Thank you very much,” I said, picking up my bags and heading towards the terminal. You know you’ve been traveling too long when a U.S. $300 bill for one day in a Renault “Junior” seems acceptable, even if only momentarily.
By the time I got to the terminal, my currency conversion process got swapped into memory and I promptly turned around and walked back to Hertz. Using a few choice words remembered from my high school days in Switzerland, along with some universally understood hand gestures, I convinced the station chief that I felt that a fee of $300 was not totally appropriate. Through some hard negotiating, I got the rate down to U.S. $125. What a bargain.