Message-Id: <9610060042.AA27741@ig.cs.utk.edu>
From: Keith Moore <moore@cs.utk.edu>
To: Eric Thomas <ERIC@vm.se.lsoft.com>
Cc: Robert Elz <kre@munnari.oz.au>, drums@cs.utk.edu, moore@cs.utk.edu
Subject: Re: cname lookup effort 
In-Reply-To: Your message of "Sat, 05 Oct 1996 19:05:55 +0200."
             <199610051716.NAA02975@CS.UTK.EDU> 
Date: Sat, 05 Oct 1996 20:42:00 -0400
Sender: moore@cs.utk.edu

> Robert, I  have described  a concrete, legitimate  and common  real world
> problem. Not  providing the users  with a solution  to that problem  is a
> non-option.  I  have shown  that  currently  CNAME  is the  only  working
> approach,  in  spite  of  clearly  unwanted side  effects. 

What you've described is a concrete, legitimate, and common real world
problem, for which neither CNAME nor MX works very well.  

A solution that might work reasonably well (for Eric's scenario, perhaps
not in general) is to allocate a separate IP address for each service, 
make sure that the host to which the service is assigned listens on 
EACH of those addresses (using interface aliases or pseudo-interfaces), 
and make sure the web server, listserv gui, etc., binds to the
designated IP address for that service.  This also has undesirable
side effects: it uses extra IP addresses and slows down packet processing
at that host, but it works okay for small numbers of addresses.

As I suggested to Eric in private mail, I don't think we're going to
get agreement to change CNAME behavior. 

I suggest that 821bis (or perhaps a separate RFC) simply document all of 
the different cases (A, MX, CNAME, and interface aliases) and the 
drawbacks of each.  I could make a strong case that it doesn't belong
in 821bis because it has more to do with DNS and other services besides
mail than it has to do with SMTP.

Keith