On Mon, Jan 10, 2000 at 09:01:49AM -0500, John R. Levine wrote:

...

Selection of Appropriate Local SMTP relAy (SALSA :)

...

An alternative is to provide a method for a mail client to discover a suitable local smart relay in some dynamic fashion. This requires some ubiquitous, global, standard database which every operator uses... ... like the DNS.

That's much too complicated. What we need are some well-known IP addresses, analogous to well-known ports, that are not routable on the global Internet, but that are assigned to standard services within each network, e.g.:

I think that both approaches have some merit, and neither are without complication. For example, the overloaded well-known-address approach might lead to operational confusion in the event that a temporary ISP doesn't provide the well-known service to a roaming user: + outbound packets from the user might spiral off into the default free zone, ultimately never connecting to anything + connection requests might follow the temporary ISP's default out to their transit provider who _does_ support the address, and that relay (correctly) refuses to relay for the user + the real identity of the well-known relay box that a given user connects to might change with time, making it harder to support the user This is going to provide operational confusion, I think. In this respect, at least, the MX idea works more cleanly in the event that there are _no_ MX records to be found. Interesting idea, though; thanks for mentioning it. It seems from some of the mail I have received that there is some interest in this idea. Rather than cluttering up NANOG any further, I'll take it to the roamops wg list at roamops@tdmx.rutgers.edu. Joe