Re: [Exim] Exim performance on large mail systems

Página superior
Eliminar este mensaje
Responder a este mensaje
Autor: Theo Schlossnagle
Fecha:  
A: EXIMUsers
Cc: exim-users
Asunto: Re: [Exim] Exim performance on large mail systems
Our machines are similar in configuration except that they are intel
machines (dual proc/1 GB). We deliver about 1.5 million message/day on
any given machine. But, we don't push mail between 11am and 8pm (only
8pm to 11am).

We run exim in -bd -q1m mode. Limit the queue runner queue_run_max to
about 50 and let it run. External to that we have process that monitors
available resources on the system and forks a queue runner if and only
if the resources are sufficient. It seems to be very stable -- we have
been running this way for about 1 year.

I figure if we ran all day long, we would probably end up delivering
about 3 million messages per day per machine -- though it is very nice
to have a maintenance window :-) The limiting factors for our machines
are disk I/O and context switching. Your Sun box should do better on
the context switching from and probably has more on chip cache which
will help there too. As for disk I/O, there are a million different
solutions. Most of them cost money though...

In light of the fact that one of our relay machines costs ~ $2000 US,
the most economical solution is to buy another machine and share the
work load that way. :-)

On Wednesday, May 2, 2001, at 05:32 AM, EXIMUsers@??? wrote:
> One of our dedicated exim mail servers, a dual cpu sun box with 1GB ram,
> currently sends about 10GB a day outbound. Recently this machine has
> peaked
> at over 800 exim processes, pushing free memory down to about 15MB.
> Are there any particular configuration changes that can help with
> performance on exim systems pushing a lot of mail and starting up a lot
> of
> proccesses? I was considering just setting queue_only, launching exim
> with
> "-bd -q1m" and then limiting the number of processes with
> queue_run_max..
> Are there any possible problems I should be aware of with this
> solution, or
> does anyone else with similar usage levels have any tricks they can
> pass on
> for optimizing the mail delivery process?


--
Theo Schlossnagle
1024D/82844984/95FD 30F1 489E 4613 F22E 491A 7E88 364C 8284 4984
2047R/33131B65/71 F7 95 64 49 76 5D BA 3D 90 B9 9F BE 27 24 E7