ARPANET is now 50 years old

I became a contributor to the ARPANET project when TCP/IP was being defined, with an interest in the problems of synchronisation and flow control in end-to-end protocols.

Starting in 1972, my work was initially focused on the first protocols designed for the ARPANET (the NCP protocol) and for the Cyclades network. I had joined the Cyclades project to create a team in Rennes (Brittany, France), where I could take advantage of a well-organised computer center. I decided on a specific work program after learning that a Simula-67 compiler was available. What about developing an event-based simulator aimed at revealing the causes of those malfunctions exhibited by the early Cyclades protocols – inspired by NCP to some extent?

That turned out to be a good idea. Simula-67 happened to be the first object-ori-ented language ever created, and programming in Simula-67 was incredibly pleasant and efficient. Simulations were focused on the transport layer (layer 4), while on the ARPANET side, they were studying the problems on layers 3 and 4. (The layering was not yet very clear.) I chose to consider only the end-to-end connections between com-puters and not the packet switching nodes (ARPANET IMPs or Cigale nodes). This made it possible to look at end-to-end synchronisation problems without being exposed to artificial limitations proper to a specific network layer.

I then got in touch with the ARPANET teams. When Louis Pouzin hired me, he men-tioned his intent to have someone from the Cyclades project immersed in one ARPANET site. This was actually an ideal prospect.

Before joining Cyclades, I had spent three years at CERN, working in an inter-national environment and sharing my office with an English colleague. I wanted to face a stiffer challenge, from scientific and societal standpoints. So I toured the ARPANET centers in spring 1973. DARPA (Washington), BBN (Cambridge, MA), Stanford University (Palo Alto, CA), and UCLA (Los Angeles, CA) were on my list. Vint Cerf had just been appointed Professor at the Stanford’s Digital Systems Lab. When I met Vint, I knew I had made my choice: top-level scientific discussions and human contact. I had brought along my simulation results. Vint told me they were extremely use-ful, since NCP was to be replaced by some other protocols (TCP and IP were in the works). So I came to Stanford in June 1973, for a year, to work on synchronisation prob-lems, flow control, error control, and abnormal end-to-end connection failures.

Vinton Cerf had just been appointed to Stanford. Did he have a team?

Yes. Shortly after obtaining his PhD at UCLA, Vint started teaching at Stanford and was awarded DARPA funding. He quickly built his team (half a dozen PhD students and one or two engineers in charge of logistics, installation of the IMP, FTP and email software).

You mention DARPA financing…

Yes, the financing was funnelled through DARPA, coming from the Department of Defense. In the late 1960s, encouraged by the work by Leonard Kleinrock (then at MIT) and Rand Corporation, DARPA launched the ARPANET project. Later on, when it became obvious that packet-switching was opening up tremendous opportunities, the National Science Foundation and the computing industry joined forces with DARPA. TCP/IP became a commodity technology in 1983, when DARPA decided to put the technology in the public domain.

Funding was military, but the project was not classified; moreover, foreign researchers were welcome among the teams…

During the ARPANET years, there was also some funding from DARPA awarded to non-US teams, which had connections to the ARPANET, the Norwegians notably. The British and the French had their own national budgets. Bringing in researchers has always been a DARPA and a US policy. To the best of my knowledge, ARPANET’s developments were not classified. There might have been some confidentiality clauses or patents related to the work conducted by BBN or a few ARPANET teams.

When you arrived at Stanford, after working at CERN and in France, did you discover a different working atmosphere?

Yes. CERN is a highly international organisation. But it is huge, and I was hired to join a team already structured, and wasn’t really asked to be innovative.

The big difference with Stanford was that over there I could really bring something new. The whole of Vint’s team was constantly brainstorming, investigating fundamental questions regarding time-varying and error-prone inter-process communication. In fact, without knowing it at that time, we were uncovering that terra incognita which went by the name of “distributed computing” years later. To some extent, we were computer scientists learning that our discipline had been so far focused on a special case, centralised systems, and that the time had come for a mental revolution akin to relativistic physics: the hypothesis according to which processes share the same space/time referential should be abandoned. In distributed systems, there is no central locus of control. Some “global state” being necessary for endow-ing any system with desired properties, specific (distributed) algorithms had to be devised to this end.

After that, your research became actually oriented towards distributed computing?

That is right. TCP was being designed, and the famous RFCs addressed fundamental questions. The foundations had been laid. A specific example comes to mind. If I remember well, it is Ray Tomlinson who, with the alternating bit protocol, proposed a primitive version of the sliding window scheme on which I was working. The hypothesis that there may be only one packet in transit had been abandoned, and we had to find a scheme whereby multiple packets in transit on a full-duplex connection could be designated unambiguously in the absence of a “natural” time/frame referential and in the presence of failures (packets had to be repeated, which raised the need to know whether any given packet had previously been received or whether it was a new one). To put it simply, network protocols in the ARPANET age were posing problems of cardinality 2 (end-to-end), whereas distributed algorithms are needed for solving problems of any cardinality greater than 2.

How were the work and research themes distributed and organised within the teams?

As far as I was concerned, besides my work on TCP, I had total freedom. For example, while in Vint’s team, I became interested in the work conducted in nearby Xerox PARC, which led to Ethernet. Also, I was reading papers on the ring developed at Irvine University (CA). Both topics had a definite influence on my work related to real-time local area networks. PhD students were offered a choice, among the themes selected by Vint. Presentations were organised on a weekly basis. Everyone picked a theme according to his desire, in connection to open problems or published papers. We were looking for solutions, and there was some kind of scientific competitive spirit in the air, as well as the conviction that we could expose any “crazy ideas and get immediate feedback. This kind of “thinking” was specific to the birth of Silicon Valley.

Is this spirit specific to the USA?

The sharing of speculative inroads is specific to research. In the USA, it was easier than elsewhere to engage massive research efforts that would attract attention, bringing in the best researchers. Awareness of emerging topics was not widespread in France at that time. When I returned to France in 1974, I started teaching a course on computer networks at the University of Rennes. In 1977, one of the professors at that university had still not understood much of it. His vision was that “issues of bits per second, message loss rates, and reliable message deliveries are not related to computing, but to physics.” Clearly, there was a huge cultural gap between French and US universities at that time.

Wasn’t it too difficult to return to France?

I was honest. I had promised Louis Pouzin I would come back to the Cyclades team, even if the temptation was great indeed to stay in the USA. We were offered golden job opportunities in those early days of emerging Silicon Valley.

How did you promote the work done in the USA?

I published a journal paper based on my work in France first and at Stanford four years after I left Stanford. I was not very interested in publishing on past work, and I was deeply invested in my research on distributed algorithms.

In the USA, I saw and learned how to conduct disruptive research: don’t follow the crowd, think by yourself, and when you are confident enough in your findings, take your responsibilities and publish/propagate, even if you break away from dominant or trendy topics.

Your contribution to the ARPANET was recognised as your name was put on the Stanford “Birth of the Internet” plaque…

Yes. Unfortunately I was unable to attend the celebration that took place on this occasion in July 2005.

Returning to France, did you operate a knowledge transfer?

Indeed. What had I brought into my nets? Layered architectures, the importance of properly separating physics (transmissions, links, computers) and virtual/logical structures (connections, processes, system models). In 1977, I published a pioneering paper on distributed computing (a distributed algorithm for fault-tolerant mutual exclusion, based on the concept of a virtual ring). I brought back information on how routing would be done in the future Internet. And knowledge relative to what would become an area of very high visibility: local area networks.

Do you remember other Europeans who worked directly within the ARPANET?

Europeans who immersed themselves in ARPANET teams? I do not remember, but there were significant contributions by Europeans. The idea of layered protocols was adopted by everyone and led to the OSI project at the International Organization for Standardization, in which the foreign community was mobilised. I had a lot of discussions with Hubert Zimmermann on the layered model,but I quickly gave up because I was more interested in my more theoretical research program. One of the members of Cyclades, Jean Le Bihan, proposed in 1978 that I should join the Sirius project he was conducting on distributed data-bases, which I did, for I could then work on distributed algorithmic issues such as concurrency control and the serializability theory.

As far as OSI was concerned, the problem was closed for me. And I have a personal bias: I hate standards committee meetings. I think they take too long to achieve convergence and, very often, good ideas come from the outside. So many would-be standards have been dis-carded only a few years after they became official! John McQuillan said something I liked very much in the 1980s: “Standards are great, there are so many you can choose from”! Standards rest on human-centric agreements, which explains why some standards are flawed, notably those resting on “solutions” that violate impos-sibility results. Current standards for inter-vehicular cooperative schemes based on wireless radio communications aimed at autonomous/automated vehicles are an example – terrestrial and aerial autonomic vehicular networks are my current area of research.

Do you have any regrets about elements that were not thought of at the time of ARPANET?

Cyberattacks, privacy threats and illegitimate cyber-espionage were not foreseen at that time. We had the firm conviction that we were developing a tool that would serve humanity, and no one really thought it could be misused. There are no encryption algorithms, no pseudonym schemes in the ARPANET protocols’ “DNA.” These issues have gained higher visibility since the advent of mobile wireless networking. We know that we are vulnerable in cyberspace via our smartphones. Matters are much more worrying with upcoming “connected” automated vehicles. With currently adopted solutions, these vehicles are equivalent to smartphones-on-wheels, making us vulnerable in the physical space: smartphones-on-wheels may kill.

If we had to conclude on what remains of ARPANET?

Gerard Le Lann: Today, everything runs on TCP/IP, Internet would not have existed without ARPANET (and external contributions, notably from France and the UK), and Internet is not dead, contrary to recurrent dogmatic “predictions”! More than anything else, packet-switching has been THE crucial revolutionary technology, experimented first with ARPANET. Without packet-switching, none of those cyber-services we rely on daily nowadays, none of those technologies that have emerged recently (e.g. IoT) would have come into existence.

In 2013, at the Ecole Militaire in Paris, after you were awarded the Wilis Lamb Prize by the French Academy of Sciences, you gave a lecture with Bob Metcalfe and Vint Cerf and I remember that there were also references made to applications in the energy sector...

Yes indeed, Bob Metcalfe had outlined the idea of a future Enernet, an open network based on energy packet-switching, where energy packets could be produced, disseminated, traded, and consumed anywhere, anytime. The future will tell whether something close to Enernet can be deployed. We have many challenges to think about. In the list of new ground-breaking technologies, one finds mobile optical communications—passive optics and active optics (visible light commu-nications). They can serve to solve fundamental problems left open with radio technol-ogies. Just one quick example: naming in spontaneous open mobile networks, where no advance knowledge of neighbours, membership, geolocations or velocities, is avail-able. Optical communications also have a decisive advantage over radio communica-tions: they can be privacy-preserving by design. As regards personal data protection, with the GDPR,29 the Europeans are ahead this time, I believe.