[Mimedefang] no free slaves...

[Jon R. Kibler]

Charles Mount wrote:
> 
> Does anyone have any wisdom on tuning an over worked server?

I agree with David's suggestion about letting MD use a tmpfs cache -- it greatly improves MD performance. 

However, we have gone even further. 
	We use a cache for sendmail's .db files (aliases, access, and virtuser). This requires mods to /etc/init.d/sendmail to load the cache on startup and requires mods to the various .db build scripts (and replacing newaliases with a script) to rebuild the .db files in /etc/mail and to then reload cache. A word of warning -- expect to occasionally see SYSERR ... cannot lockf... bad file number errors -- which result from updating the file in memory while it is being accessed. No harm done, just a tempfail occurs.
	We have cached all the AV sig files into another tmpfs.
	We have split qf and df files into separate spindles and cached xf files into another tmpfs.
	We have set the connection rate throttle (see "milter timing out" thread) and limited to total number of daemons such that we will never create more daemons than MD can create slaves.

Also, have you looked at the IO and Swap wait rates? I suspect that you do not have nearly enough RAM on these systems for the load that you are describing -- I don't know the RAM size limit for these systems, but if they were my systems, I would have AT LEAST 32GB and maybe 64GB of RAM (you can get 100% Sun compatible RAM from 3rd party suppliers for less than 1/10 of Sun's published price).

Hope this helps!

Jon R. Kibler
Chief Technical Officer
A.S.E.T., Inc.
Charleston, SC  USA
(843) 849-8214




==================================================
Filtered by: TRUSTEM.COM's Email Filtering Service
http://www.trustem.com/
No Spam. No Viruses. Just Good Clean Email.