Talk:Louis Pouzin

Datagrams

The article currently says "This network was the first actual implementation of the pure datagram model, initially imagined and described by Donald Davies", which I don't think is the best way to put it - and may have errors.

• CYCLADES was important because it was the first network to make reliability the responsibility of the hosts, not the network. This innovation was explicitly copied in TCP, as Cerf et al cheerfully acknowledge. The reference to "datagrams" is unfortunate, in my view, as it tends to disguise this key system engineering insight.

• The use of the phrase 'pure datagram model' is unfortunate, as the word 'datagram' changed its meaning over time. So one shouldn't rely on use of the word 'datagram' in a historical document to mean what it does now. It was originally a synonym for 'packet'; the meaning of 'user visible network service which is not reliable' probably started to come in with UDP. The meaning of 'synonym for packet' only disappeared slowly, after that, though.

• I'm not sure exactly how Davies' work differed from Baran's prior work, as it's been a long time since I read their stuff. I don't know whether Davies' original packet network used datagrams (in the modern sense) or not - but if it did, then the statement that CYCLADES was "the first actual implementation of the pure datagram model" is wrong. If it wasn't pure datagrams, but a mix, then Baran's work (which Davies cheerfully acknowledged the priority of) preceded him; and maybe the ARPANET too. (Whether Baran's network used 'pure datagrams' or not, I don't know - it may not have been designed/described in enough detail to make that determination.)

That's all part of why I feel that the best way to put Pouzin's contribution is to say that his design took the bold, innovative step of making reliability the responsibility of the hosts, not the network; this says clearly and crisply what the underlying engineering point is. It is this insight that led to the success of the internet model, and the Internet.

I don't have the energy to fix the article, though. There are a number of point I'd have to investigate (above) first. Perhaps someone else will. I will bring this question up on the internet-history mailing list, to get the views of those pioneers still with us on record. Noel (talk) 15:33, 27 July 2022 (UTC)