Berlin, October 5, 2009: My French is 'orrible, but with a little help I improved on Chris Edmondson-Yurkanan's translation of a portion of an interview with Louis Pouzin:
Cyclades is based on packet-switching, like the Internet. In fact, everything started with a report of Paul Baran who, in 1963, described a reliable network obtained by transmitting packets independent of one another. In 1967, Larry Roberts, then at ARPA, launched a computer network project aimed at resource sharing. Researchers could thus use remote resources from their own computer, in their home environment.
The first document on the ARPA network was published in 1968. Thus, basic architecture was known as of this date, at least on paper. The demonstration had yet to take place. At the beginning of the Cyclades project, I went to the United States and was consulting the BBN who had been working on the ARPA project. The ARPA network implemented packets in virtual circuit; our packet network, on the other hand, was "pure", with independent datagrams, as in IP today. IP took from Cyclades the "pure" packet and logical addressing rather than the physical addressing used by ARPANET. I had also introduced into Cyclades the concept of zone, which in the Internet is called domain, as well as multihoming and the concept of sliding windows for flow control. TCP/IP is thus an improved version of Cyclades. Notably, it adds the possibility of fragmenting packets along the way, a little like a train whose wheel base varies to adapt to different networks. This concept is very useful to connect local area networks such as Ethernet, which did not exist when we conceived Cyclades.