Date: Fri, 26 Mar 1999 09:38:17 +0100
To: Bart Schaefer <schaefer@zanshin.com>, Keith Moore <moore@cs.utk.edu>
CC: IETF working group on revision of mail standards <drums@cs.utk.edu>
From: Brad Knowles <blk@skynet.be>
Subject: Re: Optimising mail to many recipients
Message-Id: <19990326093817.011886@lists.skynet.be>
Mime-Version: 1.0
Content-Type: text/plain; charset=US-ASCII
Content-Transfer-Encoding: 7bit

On Thu, Mar 25, 1999, Bart Schaefer <schaefer@zanshin.com> wrote:

>(Meta-comment: should we continue this discussion on some other mailing
>list?)

    Yeah, probably.  Got any suggestions?

>By a remarkable coincidence, a colleague of mine (Bob Glickstein) was
>doing some research on exactly this topic last week.  His results:

    I was working on writing an RFC about ten years ago that would have
implemented per-domain mail routing.  Mail to moe@domain1.com,
larry@subdomain2.domain1.com, and curly@domain2.com would have sent one
copy to the MXes for .com, which would have sent one copy to the MXes for
domain1.com and one to the MXes for domain2.com, and the MXes for
domain1.com would have then forwarded on a copy to the MXes for
subdomain2.domain1.com.

    I came to realize later that this X.400-style core/border gateway
routing thing probably wouldn't work too well, since you'd have more and
more mail being dumped on the MXes for .com, and no real way for them to
recover any money to pay for that hardware.


    This would appear to be a slightly different problem, where you're
optimizing instead on sending the absolute minimum number of copies of
mail messages, regardless of where the recipients are located, what
efficient routes between you and them (and their backup MXes) may/may not
exist, etc....

    For example, if mail.uu.net (somewhere in Northern Virginia) is a
backup mail server for two hosts in two separate domains in inner outer
upper lower slobovia (which has a single 300 baud modem to Uunet serving
the entire country), then if one person on one of those hosts were to
send mail to someone on the other host and another person somewhere else
in the world, then one single copy might be sent to mail.uu.net, which
might then turn right around and send a copy down that slow line to the
recipient on the other host in the same country, as well as forwarding on
a copy to the other recipient.

    I think the lesson here is that backup MXes do not necessarily have
anything to do with packet routing efficiency, and attempting to optimize
one of these things through the other is probably a mistake.

>|                         So 1532 SMTP connections could suffice to
>|deliver those 2944 messages, a savings of 49% over the naive approach
>|of one connection per message, and a savings of 11% over the slightly
>|less naive approach of one connection per domain.

    If I'm not mistaken, this is precisely the approach taken by sendmail
and at least some other semi-intelligent MTAs.

>|                                            From my test data, it
>|looks like this extra optimization means those 2944 messages could be
>|delivered with 1384 SMTP connections---53% fewer than the naive way,
>|19% fewer than the less naive way, and 10% fewer than my first clever
>|way above.

    Since this method would only save 10% message transmissions over the
way I understand sendmail works today, I don't see how this is all that
big of a savings.

-- 
  These are my opinions -- not to be taken as official Skynet policy
 ____________________________________________________________________
|o| Brad Knowles, <blk@skynet.be>            Belgacom Skynet NV/SA |o|
|o| Systems Architect, News/mail/FTP Admin   Rue Col. Bourg, 124   |o|
|o| Phone/Fax: +32-2-706.11.11/12.49         B-1140 Brussels       |o|
|o| http://www.skynet.be                     Belgium               |o|
\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/\/
   Usenet is not the web.  Just because the web handles some things
 poorly is not a good reason to apply those same solutions to Usenet.