[ih] TCP/IP routing

[Noel Chiappa]

{Another place I need to chime in, so there's no error in the record.}

    > From: William Yeager

    > When the Internet came on line in 1983 Geoff Mogul acquired a ButterFly
    > gatewy from BBN which he named Golden. This gave the routes
    > connectivity across the to the Internet from Stanford's LAN.

'Golden' was a C Gateway, which I maintained for several years. (I have a 
sneaking suspicion that a conversation that I had with Len Bosack in the
Jacks terminal room, on one visit while doing so, later led to Cisco getting
in to the router business, but that's a story, for a different time.)

As I mentioned, a dump of the MIT-CSR machine that I have still has the
configuration files for early versions of Golden; here's a piece of one:

  #include        "../../in/in.h"
  #include        "../../nets/ether.h"
  #include        "../../nets/arpa.h"
  #include        "../../nets/arconst.h"

  /* Node name */

  char    name[]  "Golden-Gateway";

  /* Actual network configuration table. Note the trailer words in
   * the ARPANet line; for a hardware botch in the ACC board.
   * LOSERS: don't frobozz the n_max packet count fields; this
   * configuration is intended to maximize ARPANet throughput
   * in a gateway that's not computationally loaded. In particular,
   * leave the ARPANet maxip larger than the Ethernet!!!
   * "We like our defaults. If we hadn't liked them, we wouldn't
   * have made them the defaults, would we."
   */

I wound up roped into this because Jeff Mogul, who had been an undergrad at
MIT on the fringe of our group at MIT-LCS (he did a driver for the V1 ring
for the VAX VMS - there was no VAX Unix then - on the 4th floor at Tech Sq)
had gone on to grad school at Stanford. Stanford needed a router to conect
their internal internet (which was mostly - or maybe all, I was never very
familiar with their internet - 3MB Ethernets at that point). Jeff knew that
the MIT router project (the 'C Gateway') had code to talk to the ARPANET, so
they contacted us for help.


    > Bill is correct that Chaosnet did not require ARP because the host/net
    > fields in the packets were one byte each on 3mbps networks and the
    > encapsuation of the packets was one byte dest host, and one byte source
    > host.

On a 3MB Ethernet, which used 8 bit host numbers, no. (Although I didn't
realize CHAOS protocol was ever run over 3MB Ethernet; it's news to me that
it was.) CHAOS protocol on 10MB Ethernet did, of course, use ARP.

	Noel