Re: [Exim] Performance: Solaris and Linux

On Thu, 30 Oct 2003, Steven Dossett wrote:

> For the Linux Exim users out there, which file systems are you using for
> Exim?

my best experience so far is xfs with 2k blocksize, with the xfs
logdev on a separate scsi disk - i couldn't, so far, bring it to crawl
(some dell poweredge something (2600, perhaps?) with 2x2g4 xeon with
don't know what disks on a perc/3 [only used as a scsi controller,
with one-disk-per-array-stripes]). the system is not very large,
approx. 120k users, between 60-80 thousand messages per day (normal
usage pattern; when i get spammed or something, it easily doubles,
which doesn't really seem to make a difference, performancewise), but
the box also serves as a hevaily hit mysql server too. load never goes
above 0.9-1.0, usually it just sits around 0.2-0.4, depending on the
load of the sql server)

oh, and its exim 3.36 :)

> I should also mention that on both platforms I see a high ratio of CPU
> time
> spent in sys vs. user. Normal exim behavior with a lot of processes?

normal behaviour with a lot of processes - the kernel spends a
significant amount of time switching between the processes.
generally speaking, transporting mails is not really cpu-bound (unless
you do content filtering of some kind, but thats a whole different
setup), but if you have lots of processes, you are at the mercy
of the os' scheduler to behave. i don't know what the solaris kernel
does, for this kind and pattern of load i have there, stock 2.4.20
scheduler with the rmap vm does the job well. it could probably be
made better with the o(1) scheduler and some tweaks here and there,
but it just simply doesn't itch that much to worth a scratch. i, by
nature, am lazy :)

--
[-]
... and the rest is silence.