[ih] Internet-history Digest, Vol 50, Issue 6

[Noel Chiappa]

    > From:John Shoch

    > When Xerox gave Alto/Ethernet/Dover/PUP systems to a number of
    > universities (Stanford, MIT, CMU, CalTech, Rochester), I think it included
    > Nova-based PUP "gateways"

Umm, no. I cannot speak with certainty about what Stanford and CMU got, but I
can say with great certainty that all that MIT got was several UNIBUS
Experimental Ethernet cards:

  https://gunkies.org/wiki/UNIBUS_Experimental_Ethernet_interface

I believe, from scraps I heard about what happened at Stanford and CMU, that
it was the same there.

We had to write code to drive them; in the MIT AI Lab, the MIT-AI ITS machine's
CHAOS-11 got a CHAOS->EFTP protocol translator (so they could print files on
the Dover); in LCS, I added code the 'C Gateway' to both talk to the
Experimental Ethernet (we sent IP packets over it; I assume that since PARC
was on the fringes [yes, I know about the 'hints' early on] of the TCP/IP
effort, they had already allocated an Experimental Ethernet packet type for
IP, and we didn't have to ask for one), and speak PUP. (Although the only
physical network on which it was prepared to send PUPs was our sole
Experimental Ethernet - so why I wrote a PUP forwarder is a mystery to me -
there was no place to send PUPs _to_.)

(The code for both survives - I was just looking at it to give authoritative
answers about the CGW code. The CHAOS-11 code survives, too. If anyone is
curious.)

Nobody at MIT made real use of PUPs; in LCS, we had IP, and in the AI Lab,
they has CHAOS (and of course both had our own LANs, too). Dave Clark wrote a
TCP for the Alto, and IIRC it was possible to TELNET from an Alto to a TCP/IP
host. I suspect that most of the IP traffic on the Experimental Ethernet was
TFTP packets headed to the Dover Spooler (which talked TFTP on one side, and
EFTP on the other - although not via a protocol translator, I think; IIRC it
'buffered' files on disk).


    > which later led to the Stanford/Cisco multi-protocol routers)

>From what I recall of Yeager's multi-protocol router at Stanford, that was
true there, but I am not sure about MIT. (My memory is not clear on how much
of an influence PUP was; see below for more - I now think it was not much.)
At MIT, we had two competing protocol families entrenched (well entrenched,
in the case of CHAOS; less so, for TCP/IP) before Ethernet and PUP arrived,
and thus had a prior incentive to uncover the multi-protocol router approach.

There is a document from Dave Clark, "MIT Campus Network Implementation",
iniial draft dated October 1982 (which I don't have in front of me, although
I may have a copy buried in boxes somewhere), the one I do have is a later
one, from June 1983. It captures an intermediate stage in the thinking of how
multiple protocol families would be handled on the proposed MIT Campus Network.

It talks about two approaches, the first being an "MIT Standard Network
Protocol", which would be a ubiquitous packet transport service. My
recollection is that this was the only approach mentioned in the older draft
- and that I didn't think it was practical. My take came from several
previous attempts to do something like this for CHAOS and IP (the ill-fated
'MIT Protocol Word', or "Muppet"), which had utterly failed to get any
traction. So I proposed the "multi-protocol spine" approach, and argued Dave
into switching to that that approach in the later draft.

To answer charges that multi-protocol routers were impractical, I wrote
multiple forwarders for the CGW (which was started, IIRC, to show that one
could get acceptable performance from a packet switch written in a HLL; the
prevous one I did was a mind-blowing kludge written in MACRO-11, whuch used
intense macrology to instantiate all N^2 packet patha in a router with N
interfaces - code also available). Which would explain why I wrote a PUP
forwarder for it, when there was no earthly operational use for it. Dave's
later draft refers to the CGW (not by name, though - merely a 'if you think
this is not feasible, we have one working').

Dave's draft states that the protocol families the MIT Campus Network needed
to support were CHAOS, DECNet, X.25, and IP. PUP is not mentioned (except in
passing, as one of the protocols implemented on the 'see, it can be done'
machine). This makes sense; as I mentioned, except for the CHAOS-11, no
MIT machine emitted PUPs at all.

    Noel

PS: It's mildly irritating that Wikipedia credits Yeager for being "the
inventor of a packet-switched, "Ships in the Night", multiple-protocol
router". It was true independent co-invention; I don't think either one of us
had heard a word of what the other was doing.