[ih] multi-protocol routers, bridges (Was: "The First Router" on Jeopardy)

[Noel Chiappa]

    > From: Clem Cole

    > IP had not yet 'won' and a lot of site had multiple protocols running
    > on their LANs.

MIT had that problem in spades, which was exactly the genesis of the
multi-protocol router at MIT.

CHAOS had a large lead in deployment at MIT, since it got rolling before
TCP/IP really did; I think the AI Lab guys got word of what was happening at
PARC, and decided to build a rough copy (hardware and, less identically,
protcol - the latter eventually ran over DIX Ethernet too). Once LCS got
rolling with the rings and TCP/IP, it clearly made sense to have all the LANs
carry both (or we'd have to rub both kinds of wire everwhere; much more cost
effective to carry both). After Dave Clark and I chatted with Dave Moon for a
while, we came with MUPPETs (MIT Universal Packets), to allow both CHAOS and
TCP/IP packets on the MIT LANs to share a common carriage format (and thus
only one kind of router).

The CHAOS guys didn't show much movement towards actually implementhing them,
though (no surprise, not much incentive). So then it looked like carrying
each protocol independently would be easier to accomplish. That still left
the problem of inter-LAN gateways (routers), though - did those have to be
replicated? I came up with the multi-protocol router as a way to economize on
them. Dave Clark wasn't totally sold on the idea to begin with (he was in
charge of prodcing an MIT-wide campus LAN plan at the time, and included the
'multi-protocol spine' as a possible solution, but without saying 'the only
plausible solution').

So I set off to write one, to show that it was viable. Somewhere in there the
Xerox grant showed up, so we had Experimental Ethernet and PUP to deal with
too. (My first kludgy router, written in MACRO-11, handled IP packets over
the rings, the Experimental Ethernet and CHAOSNET too, but IP-only, IIRC - I
ran across the code the other day, I could look.)

(The AI guys actually implemented a service gateway to get to the Dover
printer; the AI-CHAOS-11 spoke CHAOS protocol to CHAOSNET hosts, and EFTP to
the Dover spooler. They had CHAOS protocol to NCP - later TCP - gateways too.)

Something similar for Bill Yeager at Stanford, who independently re-invented
the multi-protocol spine/router idea; PUP and IP for them.


   > Some were routable, some were not.

Interesting point, but not true of MIT (and probably Stanford too); all our
protocols were router-able. That may be part of the latter attraction of
bridges, though.

Note that bridges only _really_ work well on LANs with large, unique
interface network addresses; only after DIX Ethernet was that true. Early
LANs (Proteon rings, Omninet, etc, etc) had small addresses, and had to use
routers.

    > But mixed protocols was probably more the norm in commercials and I bet
    > University circles than not.

I'm not sure. It probably depended on what kind of boxes any particular place
had; not sure there was a commercial/academic difference, but I agree on the
'multi-protocol was more common than single rotocol'.


    > I know an early 3Com brouter but my memory is there were others. I
    > think DEC made one or two. Noel can tell us what Proeton did.

I know that at the start Proteon totally missed the bridge boat. I have to
hold up my hand for a lot of that; I was so focused on my 'a bridge looks
like a wire, but isn't' mantra (e.g. I thought we'd have to deal with
congestion explicitly - this was before Van Jacobsen), I missed out on the
advantages of bridges: trivial to install (especially with 'unique at
manufacture' interface network addresses), they supported non-routable
protocols, etc.

Also, with Proteon's own LANs not having unique interface network addresses, we
weren't well-positioned to see the ease of use, etc, points.

I think they did eventually add bridging functionality towards the end, but I
was not very involved there by then.

	Noel