[ih] History of duplicate address tests

[John Shoch]

The Experimental Ethernet ca. 1973 had 8-bit source and destination
addresses.
Depending on the machine or interface, these could have been set with DIP
switches, in software, or otherwise.
At this level there was no ability to recognize that two machines might
have the same Ethernet address.

The PUP Internet was built on top of this, ca, 1974, with 8 bits of network
number, and 8 bits of host number (conveniently matching the Ethernet).
Each source and destination PUP "port" also included a socket number, which
(for some protocols) might have helped disentangle duplicate host numbers.

This early architecture worked very well -- interconnecting thousands of
machines, dozens of networks, etc.

But this early design also highlighted some of the more complicated issues:
--An erroneously entered Ethernet host address could produce two
machines on the same Ethernet with the same address.
--Even worse, moving a machine from one network to another could yield two
machines with the same network-relative host address.

When we were finally able to publish the PUP work in the1980 IEEE paper we
already knew about the issues, and hinted at what was to be done:

"One could consider revising the entire notion of a hierarchical address
space. Under the current design, it is sometimes necessary to change the
host number of a machine which is moved from one net to another--an
operational annoyance. It is conceivable that every host could be given a
unique address within a flat address space; a more sophisticated mechanism
would then be needed to map addresses into routes, since there would no
longer be a network number as part of the address (except perhaps as a
hint, to improve performance)."
Of course, this was already underway -- 48-bit unique addresses in the
XWire/DIX/802.3 Ethernet and in the Xerox Network System (XNS) internetwork
architecture.
[Dalal published this at SIGCOMM in 1981.]

Hard to believe this was all 40-50 years ago.....

John




> Message: 4
> Date: Tue, 29 Nov 2022 15:19:46 -0800
> From: John Gilmore <gnu at toad.com>
> To: John Kristoff <jtk at dataplane.org>
> Cc: internet-history at elists.isoc.org
> Subject: Re: [ih] History of duplicate address tests
> Message-ID: <1541.1669763986 at hop.toad.com>
>
> John Kristoff via Internet-history <internet-history at elists.isoc.org>
> wrote:
> > I'm curious to know about the earliest of duplicate address detection
> > mechanisms.
>
> Were there any duplicate address detection methods used in the 3-megabit
> Experimental Ethernet?  As I recall from the early 1980s on Stanford SUN
> workstations, the 8-bit addresses used to identify nodes were configured
> by DIP switches on the interface board, so there was a significant
> chance of mistakenly configuring a duplicate address.
>
> April 1984's RFC 895 describes how IP datagrams are carried on
> Experimental Ethernets.  It provides no duplicate detection, merely
> suggesting (in those days of unlimited IP address space), "The easiest
> thing to do is to use the last eight bits of host number part of the
> Internet address as the host's address on the Experimental Ethernet.
> This is the recommended approach."
>
> Perhaps PUP or other early Ethernet protocols did duplicate address
> detection?
>
>         John
>