On Thu, Sep 07, 2000 at 10:36:38AM -0500, Ken Jones wrote:
[snip]
>
> I've had good luck with some rate limiting tricks.
>
> 1) Inject 500 messages then wait 2 seconds. This seems to give
> qmail-send time to process the 500 messages and kick off more
> qmail-remotes before it has to deal with the next 500 new messages.
This is the problem sometimes referred to as 'thundering herds', because
*one* process is handling the move from todo/ to local/ and/or remote/,
*and* is spawning new deliveries.
> 2) Call qmail-queue directly, saves on one fork/exec
Hardly relevant savings :)
> 3) have the perl script monitor the size of the queue. Slow down
> the rate of injection if the queue gets above 10K or if the
> todo gets above 2K. Qmail's ability to fork qmail-remote
> processes drops when the inode trees get large.
That's what conf-split helps with :)
The real trick to high mailinglist performance is only injecting a
message once. qmail is excellent at high-rate delivery of one message to
20.000 recipients. It sucks at handling 20.000 separate messages all
injected at the same time.
Greetz, Peter
--
dataloss networks
