On Wed 25 Jan 2006 11:34:10 GMT , Peter McEvoy <pete@???> wrote: > I had similar issues, the solution, as someone on this list kindly
> explained, was to limit the size of messages that SA scans.
Also do some fairly basic sysadmin operations to (a) determine what the
cause of the overload is, and (b) try to fix it:
1. Check swap usage. If you're thrashing swap, you need more physical memory.
2. Check disk IO using vmstat, or iostat from the sysstat package. If
your hammering your disks on one specific partition - usually /var -
consider mounting that filesystem (if it's EXT3) with the "noatime"
option [0]. Alternatively, consider buying more spindles and creating a
fast RAID array, then use that for your Exim spool partition.
3. Consider using hdparm to tune your disks if they are IDE/PATA.
Serial ATA, most hardware RAID and SCSI disks won't work with this, or
have any benefit.
4. If you're still having problems, consider putting your Exim hints
databases and Bayes databases onto a ramdisk. If you need persistence
across reboots, ensure that Exim and/or spamassassin's init script does
some clever archival when the processes startup and shutdown.
If none of that makes a difference, throw hardware at it.
Hope that helps
Graeme
[0] "mount -o remount,noatime /var" while the system is running should
see you achieve this without much pain. Make sure you add "noatime" to
/etc/fstab in the right place to make this persist across reboots.
