From: Robert Elz <kre@munnari.oz.au>
To: Keith Moore <moore@cs.utk.edu>
Cc: Matt Curtin <cmcurtin@interhack.net>,
        "Larry Osterman (Exchange)" <larryo@exchange.microsoft.com>,
        Pete Resnick <presnick@qualcomm.com>, DRUMS <drums@cs.utk.edu>
Subject: Re: Message-ID (was: RE: Take 2: References/In-Reply-To replaceme nt text) 
In-Reply-To: Your message of "Wed, 02 Jun 1999 09:35:28 -0400."
             <199906021335.JAA01766@astro.cs.utk.edu> 
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Date: Thu, 03 Jun 1999 00:32:17 +1000
Message-Id: <5163.928333937@munnari.OZ.AU>
Sender: kre@munnari.oz.au

    Date:        Wed, 02 Jun 1999 09:35:28 -0400
    From:        Keith Moore <moore@cs.utk.edu>
    Message-ID:  <199906021335.JAA01766@astro.cs.utk.edu>

  | suppose we were to define a message-id.net domain (haven't checked
  | to see if it's taken yet...) and then allow anyone in the world to
  | create message-ids of the form UUID@message-id.net, where UUID is
  | a hex representation of a locally-generated UUID   would that
  | satisfy your criteria?

That would be fine - just as long as that domain name has a particular
defined algorithm for message-id generation which guarantees unique
messages ids (where "guarantees" means "makes it very likely in normal
circumstances") no matter where they're generated.   The microsoft algorithm
as described here soulds like one where that would be true, and hence a
static fixed domain name on the RHS of all message-id's it generates would
be fine (just provided it is a fixed domain name that is properly their
perogative to add).

  | and how would this be better than using
  | a non-fqdn as the domain, say UUID@uuid (the latter "uuid" being
  | a literal string)?

No difference at all, provided that "uuid" is defined that way, and is
defined so that if it is used, no-one else can ever use it any other way.
That more or les amounts to reserving the TLD "uuid" for this purpose,
exactly as has been done with "MHS".

If that isn't done, then anyone can stick "@uuid" on the end of any message-id
(and quite properly, violating no standards or conventions at all) and all
hope of uniqueness is gone completely.

Using the longer domain name seems like it is simpler, more flexible, and
less prone to political bunfights all around.

But whatever we do, the message-id RHS *must* serve to disambiguate the
algorithm used to generate the unique string n the RHS.   Exactly how the two
work together is flexible, but for this to work, one of them (at least) needs
to use a defined assignment method (everyone generate a random string doesn't 
work).   Using the DNS, which already has all themechanism we need, seems like
the simple solution - it is also the solution that is used now, and has been
used for many years (since back before the DNS existed, and it was the NIC 
HOSTS.TXT file that was used).

kre