Bryan White <[EMAIL PROTECTED]> writes on 23 August 2000 at 09:35:55 -0400
> > Just out of curiosity: is there an obvious reason that I'm missing that
> > a custom app would send out mail faster than qmail?
>
> This is an experment, I don't know what effect it will have.
It's been frequently discussed, and the gurus think it sounds like the
winning strategy for mass delivery of different messages. I agree
with them.
> My
> understanding is that currently:
> 1) My Program fork/execs qmail-inject and sends it the letter over a
> pipe
> 2) qmail-inject fork/execs qmail-queue and sends it the letter over a
> pipe
> 3) qmail-queue writes the letter to disk
> 4) qmail-send sees the message on disk and sends a command to the
> running qmail-rspawn.
> 5) qmail-rspawn fork/execs qmail-remote and reads the letter from disk
> and sends the it over a pipe
> 6) qmail-remote sends the message
>
> I have no doubt that qmail is very robust because of this mechanism and I
> fully intend on relying on it to handle anything that fails my first
> attempt.
And you also noticed both the number of forks, and the number of
filesystem synchs (and hence actual physical disk IO) required in this
process, I guess.
The elegant approach seems to me to be to use qmqpc to queue the
messages needing queueing, probably on your normal mail server (and
run the generation and direct sending on dedicated boxes).
--
Photos: http://dd-b.lighthunters.net/ Minicon: http://www.mnstf.org/minicon
Bookworms: http://ouroboros.demesne.com/ SF: http://www.dd-b.net/dd-b
David Dyer-Bennet / Welcome to the future! / [EMAIL PROTECTED]
