Date: 6 Mar 1996 07:31:59 GMT
Message-ID: <19960306073159.13521.qmail@koobera.math.uic.edu>
From: djb@koobera.math.uic.edu (D. J. Bernstein)
To: drums@cs.utk.edu
Subject: Re: re PIPELINING

> 	(1)  require N processes with most existing MTAs,
> 	(2)  require N times the CPU cycles on the server,
> 	(3)  require N times the network bandwidth,
> 	(4)  require N times the storage for the queued
> 	     messages as they are received by the MTA,

These are _insignificant_.

If you go through your logs you will find that the _maximum possible_
benefit of multiple RCPTs for your host is under 1% of your CPU time,
under 1% of your processes, under 1% of your network bandwidth, and
under 1% of your storage, compared to your normal resource use for mail.

Latency is different. EVERY SINGLE MESSAGE has to have low latency.
It is completely unacceptable to a user that he received an important
message six hours late because he was at the end of a long mailing list.
Even if this happens for only one message out of every ten thousand,
it's a disaster.

Sometimes the disaster was somebody else's fault---a host went down---
but if it's your mailer's fault then your mailer is poorly engineered.

> (5)  lose the ability to eliminate duplicate messages

Wrong. This is trivially handled by the LDA---where you get the added
benefit of applying it to messages that came in through different
paths, messages that were duplicated because of network burps, etc.

> (6)  lose the ability to optimize storage of the

Wrong. This is trivially handled by the LDA---where, once again, you
get to apply it to messages that came in through different paths.

---Dan