--On 20 November 2008 10:34:55 -0800 Brent Jones <brent@???>
wrote:
> I have a small load server (70k deliveries a day) and thought I'd try
> using queue_only, with queue_run_max = 15 to see if it makes things a
> bit more efficient.
It won't. Exim is designed to deliver immediately - for efficiency reasons.
Queuing isn't as efficient as immediate delivery, unless you've got a
poorly connected server.
If you have multiple queue runners (eg by using -q1m as you say later),
then you'll have multiple queue runners fighting over the queue.
I have a cluster of machines that each handle similar loads to yours. More
than 95% of messages are delivered within two seconds. The other 5% are
mostly delayed by anti-spam mechanisms on remote servers - like greylisting
or teergrubing.
> I've discovered it will only spawn about 5 queue runners, and messages
> will queue up for up to 15 minutes before a runner finally gets to it.
> I have split_spool_directory enabled, load on the server is 0.05, disk
> IO is negligible, I don't see why Exim spawns so few queue runners and
> why they take so long to process messages.
> It performs flawlessly when it is not in queue_only, queue time and
> delivery time is ~1 second, and the only messages in the queue are
> frozen or in retry state (about 100 messages on average).
So, why bother with queue_only.
> Here is so relevant lines from my config, Exim 4.68, FreeBSD 6.2, Quad
> Cores with 4GB RAM and 15k SAS disks.
>
> split_spool_directory
> queue_only
> queue_run_max = 20
>
> remote_max_parallel = 10
> ignore_bounce_errors_after = 4d
>
> timeout_frozen_after = 4d
>
> delay_warning = 2h:8h:24h
>
> smtp_accept_max = 1000
> smtp_accept_max_per_connection = 4000
> smtp_accept_queue_per_connection = 4000
> smtp_accept_max_per_host = 200
> smtp_accept_reserve = 40
> smtp_reserve_hosts = obscan_hosts
> message_size_limit = 50M
>
> Any thoughts?
>
> --
> Brent Jones
> brent@???
--
Ian Eiloart
IT Services, University of Sussex
x3148
