Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Posting-Version: version B 2.10.2 9/18/84 SMI; site sun.uucp
Path: utzoo!watmath!clyde!burl!ulysses!mhuxr!mhuxt!houxm!mtuxo!mtunh!mtung!mtunf!ariel!vax135!petsd!pesnta!amd!amdcad!decwrl!sun!guy
From: guy@sun.uucp (Guy Harris)
Newsgroups: net.mail.headers
Subject: Re: RFC920 domains
Message-ID: <2373@sun.uucp>
Date: Tue, 2-Jul-85 02:47:37 EDT
Article-I.D.: sun.2373
Posted: Tue Jul  2 02:47:37 1985
Date-Received: Thu, 4-Jul-85 04:58:05 EDT
References: <918@sdcsvax.UUCP> <532@deepthot.UUCP>
Organization: Sun Microsystems, Inc.
Lines: 48

> Domain naming schemes relate to routing as follows:  If you don't know
> how to route a message directly, you FORWARD the message to someone that
> you hope will know better.  In general this means a server for the given
> domain.

Do you mean "network" (i.e., ARPANET/Internet, UUCP net, CSNET, etc.) or
"domain"?  A lot of the confusion in this debate is caused by thinking that
there is a one-to-one correspondence between networks and domains.  This
need not be the case.  Some systems assume that there is such a
correspondence (i.e., some "sendmail.cf" files use a UUCP route generator
and mailer if they see a host name that ends with ".UUCP", and use an
ARPANET SMTP mail connection if they see a host name that ends with ".ARPA"
- or route mail to a host with an ARPANET connection if they see ".ARPA"),
but a host in domain ".COM" may be on the UUCP network, the ARPANET, or some
private network.  You can do routing by "brute force" by having a huge
database of hosts and routes to them, or you could send all your outgoing
mail to a "big brother" who does the routing for you, or...

Think of name-to-(best)-route as a function (if a host has two or more
equally good routes to it, pick one randomly).  The value of a function from
one finite set to another for a particular member of its domain can be
computed by table lookup (i.e., the route database) or by "cleverer"
techniques.  Encoding part of the value in the input (i.e., using network
names as domain names) is one such technique.  However, this puts
constraints on the use of domains; domains were intended as a way to
simplify the administration of a host namespace, not as a way of cataloging
all electronic mail networks.

You can debate whether the UUCP network should be a domain with subdomains
assigned for routing convenience, but it's not a foregone conclusion that
this is the best way to go.

> The problem lies in making sure that you are really getting closer.  That
> is, that the host you forward to has a *better* idea (or at least equally
> good) of where to pass the message to.  This way, routing can be done just
> one hop at a time.
> 
> It doesn't matter whether you use geographic regions, corporate
> affiliations, or any other scheme to create the subdomains.  Everyone
> shouldn't have to know the route to everybody else and those that don't
> should use the domain structures for partial routing.

If you choose "any other scheme" to create the subdomains, you can choose
one where the domains are set up for administrative convenience, not routing
convenience, in which case there's no way to tell what a "closer" host might
be simply by looking at the components of the target host's name.

	Guy Harris