Re: The UCLA 360/91 on the ARPAnet/Internet
Michael Greenwald <mbgreen <at> seas.upenn.edu>
2012-05-13 16:41:59 GMT
On 5/13/12 6:58 AM, Noel Chiappa wrote:
> > From: Vint Cerf<vint <at> google.com>
> > I don't think we had a name for the group of implementors sponsored by
> > ARPA.
> I think at the time we informally called it the 'Internet working group' - I
> didn't realize at the time that that was a duplication of the name of the
> earlier international high-level design effort! (No doubt this will confuse
> unwary historians in the future! I've seen it referred to in IEN's that
> way too (e.g. IEN-3, IEN-191).
> Just for additional confusion, at that point we referred to the Internet
> Protocol as "IN" (see IEN-53 for an example of this), not "IP".
> > Dave Clark did his IBM PC version probably around 1980?
> Dave didn't do the PC version - that was John Romkey and Dave Bridgham. Dave
> had worked on the Multics TCP to start with (I'm not sure who wrote most of
Just to clarify in the wealth of 'Dave's: Dave
Clark worked on Multics IP/TCP and not Dave
Bridgham (as far as I remember).
> it - see:
> for more - I think Michael Greenwald took over maintaining on it after Dave,
> and then Charlie Hornig);
Yes. We actually split the IP stack; I'm not 100%
sure of my memory (I wish I had my multics email),
but I think that by 1979-80 Charlie was handling
TCP and I was handling everything else (IP, the
Internet Imp daemon), GGP (Multics actually ran
experimentally as a gateway), TFTP, UDP, the mail
protocol (I think I wrote an MTP implementation
before the SMTP spec was written? Although, I
think SMTP was a set of minor modifications to
MTP), ICMP when it turned up, TELNET (I think I
needed to rewrite it to handle some new option
negotiation, and interact with echo negotiation to
make emacs work) -- I'm not sure of the dates of
any of those). After Charlie left we did a few
rewrites of TCP and IP to reduce the number of
context switches, to deal with congestion, and
then a rewrite to have the IP stack run in an
inner ring. Dave Vinograd at Honeywell CISL hired
me to come over and turn it into a product. I
worked on this transfer for a couple of months, I
am pretty sure that J Spencer Love took over from
me there, and did all of the hard work.
I'm not sure if this level of detail is useful to
anyone, or just noise.
> he then did one in BCPL for Tripos while on
> sabatical at Cambridge, and after he got back, and MIT got the Xerox
> Alto/Dover donation, he moved that one to the Alto.
Double check with Dave. I think the Alto donation
occurred well before his stay at Cambridge, and I
think we had an Alto TFTP implementation well
before he left. I don't remember if Dave's
original small TCP/Telnet client on the Alto
predated his Tripos implementation --- I have a
vague memory that he rewrote the Alto
implementation when he came back, but I may be
totally off there -- he may have written it from
> Some of the ideas he did on his "user TELNET centric TCP" were used in a TCP
> we did for PDP-11 Unix (by Larry Allen - Liza Martin did a more classical TCP
> for the same machine - both used the same kernel support, where only the
> packet de-mux was in the kernel, and the rest of the TCP was in the user
> space). I then did a TCP for Bridge based on the Allen one. The IBM PC one
> came after all those, I think (maybe not after the Bridge one, but definitely
> after the others).
> I don't have dates off the top of my head for much of that (although I could
> probably research it in my archives if anyone needs to know); I can say for
> sure that the Alto one was done around March 1980.
> > Danny Cohen and David Reed were proponents of splitting off IP but I
> > don't think they were on the ICCB
> I'm pretty sure Dave didn't. He phased out of network work shortly after
> doing UDP (January 1979). I don't recall what he switched to working on (I
> had thought it was his PhD thesis, but I see that was done by '78).
Dave Reed was still working on networking, in
addition to Swallow, after 79. He worked on what
he called "non-FIFO protocols", BLAST (a
disk-to-disk transfer protocol, predating NETBLT,
but influencing it) and Blink (a remote display
protocol that was contemporary with Bob Scheiffler
starting X down the hall, but we thought
(incorrectly, it turns out) that the sequencing
imposed by TCP would make it too slow for running
the graphic displays for our operating system
(Swift)). This doesn't say anything about whether
Dave Reed was on the ICCB (but you can ask him).