Re: [Exim] maximizing mail thruput ?

> > Or can I just increase the load factor to 100 and let it ride ?
>
> With a load that high, your machine is just not going to do anything useful
> really (unless it has got 50 processors <grin>).

That's not entirely correct. Load is the average number of runnable
processes. For one, the question is what counts as runnable. Processes
that wait for short term I/O or page faults to be completed? More CPUs
don't help for that. Then load will arise with increased burstiness
of runnable jobs. There is nothing you can do against that, unless you
can influence your runnable jobs, but neither something to worry about.

I suggest to check how much CPU time is spent for what. "vmstat 1"
gives you an idea what the system does. I rarely have a CPU problem,
disk latency (not bandwidth) is my usual bottleneck. Using the highest
possible SCSI tagged command queue size helps a lot. On Linux, e.g. the
SYMBIOS/NCR drivers use only 8 tagged commands by default, which is way
too low.

Load without any other numbers really doesn't tell much. It's just used
so often because it is easy to obtain. I suggest to use vmstat/iostat
and eximon for interactive purposes and MRTG for non-interactive disk
and ethernet I/O monitoring as well as load and queue size. As long
as I/O throughput, in particular ethernet, keeps increasing while the
queue size does not grow too bad (a certain fixed queue size is ok and
to be expected under heavy traffic), don't worry about load and set your
load border somewhat beyond the amount where I/O starts to decrease and
mails queue up badly.

It's not a great solution, but good enough for daily operation.

Michael