Re: [Exim] Performance question

On Nov 11 Michael Brutman wrote:

>My comments on kernel mode were based on profiling data that I collected.
>The profiling data showed the time to be in process/thread creation, not
>I/O.

OK.. maybe you want to build the thinnest Exim you can for that bit. Does
compiling with "-Os" (and obviously excluding all the transports/lookups
etc. you don't need) help?

>I know that exim is nasty on the disks. I've run tests with EXT3 and
>Reiser, with mixed results. Both of those are journalling file systems so
>I expect there to be more writes than usual. What is recommended ? Does
>it make sense to run EXT2? (I plan to measure and find out.)

I think ext3 with data=journal, an external journal, or maybe both!
Try RAID 0+1, or RAID 10 in preference to RAID 5.

You'll probably find significant I/O improvements in Linux 2.5, if you're
feeling suitably scary. In particular ext3 should take finer locks (which
should help your SMP system performance).

Please let us know what makes it faster for you! Testing with small
messages should help iron out such bottlenecks.

Matt