On Fri, 18 Aug 2006, Philip Hazel wrote:
>
> In the case of queue_only_load, I think it might be hard to pick a
> number that would be suitable for the huge variety of hardware
> configurations that Exim runs on. Maybe. I'm no expert in exactly what
> the load value means.
Traditionally it's the average number of CPU-limited and disk-limited
processes. The right load limit setting depends a lot on what your
workload is. If your main bottleneck is disk, then you don't want your
load to get up near one. (You need some headroom to recover from bursts.)
If your main bottleneck is CPU then loads up to the number of cores are
OK.
So in the situation where a lot of messages are being pumped down the same
connection, a lot of delivery processes will be fired off concurrently,
which implies a spike in disk load. Queue runner processes, on the other
hand, run serially, so are smoother on the load. For stuff like
SpamAssassin, you can manage the imposed load by tuning the number of
spamd children to some small multiple of the number of CPUs (allowing for
network test latencies), with a suitable connection backlog limit to allow
for burstiness.
Tony.
--
<fanf@???> <dot@???>
http://dotat.at/ ${sg{\N${sg{\
N\}{([^N]*)(.)(.)(.*)}{\$1\$3\$2\$1\$3\n\$2\$3\$4\$3\n\$3\$2\$4}}\
\N}{([^N]*)(.)(.)(.*)}{\$1\$3\$2\$1\$3\n\$2\$3\$4\$3\n\$3\$2\$4}}
