Subject: Re: Mail list envelope sender address
To: None <current-users@NetBSD.ORG>
From: D. J. Bernstein <>
List: current-users
Date: 11/25/1996 22:09:57
Perry Metzger writes:
> Look, as much as I think its fun pointing out how many ways Dan
> justifies his lazyness as good design,

Metzger is stunningly ignorant of how mail works.

He continues to claim that parallelism does not reduce latency.

He continues to claim that there is no security problem with throwing
out messages that have duplicate Message-IDs.

He continues to claim that sendmail never did just-send-8---even though
I've pointed out to him that sendmail did just-send-8 starting in
version 6.57. (See the sendmail RELEASE_NOTES.)

> the completely bogus proposition that this is a performance enhancement

qmail #1: open, wait, helo, wait, mail, wait, rcpt, wait, data
qmail #2: open, wait, helo, wait, mail, wait, rcpt, wait, data
sendmail: open, wait, helo, wait, mail, wait, rcpt, wait, rcpt, wait, data

There's an extra RTT for sendmail. Metzger's handwaving won't make it
disappear. (PIPELINING would, but most servers don't support that.)
Measured sendmail slowdowns range from 5% to 80% for two recipients.

Sometimes there are several recipients. Sometimes a RCPT response takes
several minutes. The result is that, on occasion, a message is delayed
by more than an hour on a single hop. That's unacceptable.

> no, Dan, I don't believe you no matter how many graphs you come out with

Some people don't profile code because they aren't worried about speed.

Metzger doesn't profile because he doesn't believe the profiler.

> ...this is the NetBSD mailing list, not the qmail mailing list.

Metzger and Fair are allowed to criticize qmail, and I can't respond?

Anyway, if people address further questions to comp.mail.misc, I'd be
happy to answer them there.

Sick of sendmail? Don't get mad; get qmail.