Date: Mon, 18 Mar 1996 09:46:53 -0800 (PST)
From: Mark Crispin <MRC@Panda.COM>
Sender: Mark Crispin <mrc@Ikkoku-Kan.Panda.COM>
Subject: re: multiple RCPTs, again
To: drums@cs.utk.edu
In-Reply-To: <19960318085126.21748.qmail@koobera.math.uic.edu>
Message-Id: <MailManager.827171213.3608.mrc@Ikkoku-Kan.Panda.COM>
Mime-Version: 1.0
Content-Type: TEXT/PLAIN; CHARSET=US-ASCII

On 18 Mar 1996 08:51:26 GMT, D. J. Bernstein wrote:
> It is, of course, patently obvious what ``extreme circumstances'' are:
> sending an large message over a slow link to multiple recipients at a
> host very close by.

There is no requirement that the message be "large" (although it bewilders me
that anyone can consider 10K to be "large" in this day and age).  A large
message (at least 100K) would certainly exacerbate the problem, but that
doesn't mean that it doesn't exist with "small" messages.

Bernstein's example of a "small" message (3K) still needs 2 seconds on a
V.32bis line.  That time can not be multiplexed.  It can be compressed by
V.42bis, but nonetheless it is time when 100% of the bandwidth is used by that
data and that data alone.  Any attempt to mix other data (e.g. alternate
datagrams with other streams) will make things worse because then TCP header
compression gets munged.

There is no requirement that the link be "slow".  Any server which fails to
trim the number of simultaneous incoming SMTP connections from a single client
(common) and which also has a resource limit on the number of simultaneous
incoming SMTP connections (also common) will run into the situation in which a
single system can monopolize all of its incoming SMTP connections.  This
happens often enough by accident, without deliberately trying to make it
happen.

There is no requirement that the host be "very close by".  The host could be
very far away.  I can tell you what happens when you have 50 SMTP streams all
trying to deliver (different) messages to Brazil.

> Now let's look at Mark's example: sending a 10KB message over a 14.4k
> link to multiple recipients at a host that responds within 0.2s,
> including RTT. All of his errors aside, he managed to correctly observe
> that separate SMTP connections aren't a win in this case. So what?

Gee, it sure seems to be that there are a lot of such sites out there.  There
also are a lot of mail hubs out there.  These are not "extreme circumstances"
although they may be outside of Bernstein's experience.  The fact that certain
circumstances may be outside of someone's experience is why we have working
groups instead of just having one person decide everything.  But working
groups don't work with close-minded individuals who won't listen to
information which is contrary to their own expectations.

It is curious that Bernstein presumes to use such phrasing as "all of his
errors" in reference to someone who's been part of Internet mail standards
development for nearly 20 years.  Several other folks here have similar
experience.  The term "sophomoric" is applicable to someone who vehemently
insists that he is right in the face of those with more experience and first
hand knowledge.

Once again, if Bernstein expects others to listen to his claims he needs to
pay attention to what others have to say.  He needs to accept that just maybe
someone else may have a better idea, or may know something that he doesn't
know, or work in environments in which he doesn't work.  It's fine to make
claims without taking responsibility for what gets unleashed by acting on
those claims.