[ih] Significant milestones in the history of TCP/IP

[Noel Chiappa]

    > From: Vint Cerf <vint at google.com>

    > actually their experience with a one-node local area network influenced
    > Roberts' choice of data rate. Roger Scantlebury attended the 1967
    > meeting at which he met Larry and said that higher speed would reduce
    > delay. Larry ended up with 50 Kb/s lines rather slower 2.4 kb/s lines.

Ah, right, now that you mention it, I do recall that. I wonder how important
a technical point that was, though? (See discussion below.)

I should emphasize that I'm mostly focused on technical influences, in asking
this. I know he and his co-workers had a big influence, in terms of 'moral
suppport', etc, but that's different. I should further emphasize that I have
_no_ axe _at all_ to grind, here - I simply want to know the truth. If he
deserves a ton of credit, fine; if his _technical_ influence was modest, fine
too. (And as a UK citizen living in the US, I claim to be pretty free of
national bias! :-)


    > From: Alex McKenzie <amckenzie3 at yahoo.com>

    > Davies team told Larry Roberts about Paul Baran's work, which Larry was
    > unaware of.

I think a better way to put it would be that they got Larry to go back and
look at Baran's work, because the documentary record shows _definitively_
that he had contact with Baran's work _before_ he met Roger Scantlebury at
the October, 1967 meeting in Gatlinburg.

If you look at Baran's Oral History from the CBM (OH-182), pg. 37, you will
find that Baran found documentary records (his old calendar) showing a visit
from Larry Roberts in February, 1967 (the 28th, to be exact) - well before
the Gatlinburg meeting.

The interesting question, of course, is why Larry _apparently_ didn't pay
much attention to Baran's work until the NPL people mentioned it? This is
probably related to a question I've asked here before - why Baran's paper in
the '64 IEEE ToN journal didn't get more attention. Maybe Larry got too
focused on the _goals_ of Baran's work, and thus felt it wasn't relevant to
his project?


    > At a later time Davies team (Roger Scantlebury) also convinced Larry to
    > build the ARPAnet with a small number of "high speed" (50kbps) lines
    > rather than a large number of "low speed" (9.6 & 19.2 kbps) lines.

I'm trying to work out how important that was, technically. Would an ARPANET
built, as you put it, of more, slower lines have worked? I.e. how critical
was that intervention? We can only guess as to how the history would have
unfolded without it, although one can do some analysis.

The routing in a network with more, slower, links would have become an issue
sooner than it did - becaue of both the size of the routing table, and also
the slower update rate on the slower lines. However, but it's unclear how
much of a problem that would have been - lagging response to changes was,
after all, eventually solved by McQuillan. Would the network have behaved so
poorly that it didn't last long enough for McQuillan to show up?

I'm not sure about congestion, etc (e.g. re-assembly) issues - would those
have been enough worse in an ARPANET with many more, but slower, links?

And of course the big un-knowable is 'had the network been attempted to be
built with many, slow, links, and had it not worked, would the people
building it have recognized the need to move to faster links, and if so, how
soon'?

I ask these questions (even though they have no immediately clear answers -
or for the last, probably cannot be answered) because the answers to them are
needed before one can really assess how important a technical intervention
this was.


    > Whether anything about the ARPAnet had much to do with the history of
    > TCP/IP is a different question, but if people think the answer to that
    > question is "yes"

I'd say very much 'yes': it showed that packet switching could work, taught
us a first round of technical lessons in that area, and many of the
applications were staples of the early Internet.

Of course, in an alternate history, maybe NPL's work would have done the same
if the ARPANET never existed, so who knows? Perhaps the CYCLADES work was even
more important, because I don't know of an alternative that might have taught
us the lessons it did about moving reliability into the hosts.

       Noel