[ih] Separation of TCP and IP

[Noel Chiappa]

    > From: Craig Partridge

    > I looked at this question twenty (?) years ago and swapped notes with Dave
    > Reed and I think Danny Cohen.

Did that all get written up anywhere?

When someone in the future wants to look at the history of the development of
TCP in some detail (i.e. an enginerring historian), there wil be pretty good
sources: various TCP draft versions; meeting minutes; etc.

The TCP/IP split ... not so much.

    > Yes Danny was doing voice experiments and TCP wasn't cutting
    > it. There was a hallway discussion of Jon Postel, Danny and Dave (and I
    > think that's the list) in early '78 (so you've correctly dated it) and
    > the agreement was to split TCP in two and create UDP.

I looked at some history books (one popular; one semi-technical) and they
also mentioned this, but gave no details of the technical impetus that drove
this. Hafner covers it on pg. 236; she also mentions the PARC people. (I
noted one un-related minor error; she has NCP as 'Network Control Protocol'.)

Abbate has the same discussion; pg.130. She drops a bad clanger there,
though; she says it made routers (still called gateways then) simpler, which
I am pretty sure is not correct; I'm pretty sure pre-separate-IP routers
didn't know anything about TCP connections. Only the code could verify that,
and I'm not sure it has survived. I do have some BBN BCPL router code; I'll
have to look at it, and see if I can work out which version it is for.

Salus doesn't seem to mention the split.

   > Key point is splitting TCP and creating UDP happened at the same time.

The general concept of UDP was I expect worked out then, but the protocol
spec (IEN-71) didn't appear until January, 1979; quite a long time after the
V4 headers first appeared - in IEN-44, June 1978. Amusingly, Jon did a host
name lookup quite a while after the V4 headers appeared - IEN-61, October
1978, "Internet Name Server" - but it runs directly on top of IP, no hint of
UDP! (Although he does say "the internet header protocol field should carry
the value indicating raw datagram. An additional field is needed somewhere to
multiplex the various applications using raw datagrams."


    > From: Scott Bradner <sob at sobco.com>

    > a story is that Vint, Jon, Danny, David Reed and someone else got
    > together in a hallway at ISI and convinced Vint to add the unreliable
    > option (what became UDP)

    > Vint says he does not recall that meeting but David Reed told me that
    > it happened and the driver was Danny's want to do voice

It's good to have that on record.

What started this is that I was looking at Danny's Wikipedia page, to find
out what name he was under, with a view to adding him to the list of notable
Technion alumni. I was astonished to see that it gave (effectively) as his
main claim to fame as the Prof. Finnegan writings! (Probably from some Gen
Xer for whom something doesn't exist unless it's online.) They are amusing
and pithy, but not - by a long way - his most important technical legacy. I
have updated the page, and will add cites to Hafner, and this discussion.

	Noel