Path: utzoo!utgpu!watmath!clyde!att!osu-cis!tut.cis.ohio-state.edu!bloom-beacon!bu-cs!purdue!decwrl!jumbo!murray
From: murray@jumbo.dec.com (Hal Murray)
Newsgroups: comp.protocols.tcp-ip
Subject: Re: ToasterNet (was Re: Running out of Internet addresses?)
Summary: Globally unique host IDs avoid installation confusion
Message-ID: <13469@jumbo.dec.com>
Date: 13 Dec 88 09:15:24 GMT
References: <8812121420.AA27357@mitre.arpa>
Organization: DEC Systems Research Center, Palo Alto
Lines: 23

The idea behind a globally unique 48 bit ID was to avoid the confusion
that results when somebody sets the switches wrong. The Altos on the early
3MB ethernet had switches/jumpers for the host number. Every now and then
somebody would move a machine from one building to another and plug it
in to a handy drop cable without checking in with the local host number
czar.

Now, suppose that host number is already assigned to another machine.
Suppose one of the overlaping machines is talking to a server. The ack
goes back, and both machines hear it. The wrong one doesn't know about the
socket, so it sends a reject to the server. The server closes the
connection. The next packet from the first machine to the server hits a
dead connection. Soon, the user gets a strange error message.

The Altos normally ran with the ethernet interface disabled. It was only
activated when you ran a program that used the ethernet, say to print
something or FTP a file. That meant that this sort of confusion was likely
to be very mysterious because the time when things actually acted up was
long after the "move" that caused it.

It makes more sense in the XNS protocols where the 48 bit ID is used
directly as the host number. That way you don't need a parameter file for
the ARP info. (They added 32 more bits for the network number.)