Re: OpenLDAP and BIG mail environments

["Kirk A. Turner-Rustin" <ktrustin@owu.edu>]

On Mon, 16 Feb 2004, Micah Anderson wrote:

> In my never ending search to try to get my performance back to normal,
> I implemented some of the slapd.conf changes that Quanah indicated
> were working in his environment.
> 
> I moved my threads down from 64 to 32
> 
> I set my idletimeout to 30 (it wasn't set)
> 
> I added checkpoint 1024 5
> 
> I brought my cachesize up to 10000 (from 2000)

Forgive me if I've missed this somewhere along in this thread,
but (assuming that the above is the setting of your entry cache in
slapd.conf), have you properly sized your BDB set_cachesize in DB_CONFIG?

If in doubt, I highly recommend this post from Howard Chu:

http://www.openldap.org/lists/openldap-software/200311/msg00469.html

I calculated our minimum cachesize based on his suggestions, then
started adding cache in 1MB increments to goose the hit rate to 100%
and the clean pages forced from cache as close to zero as was reasonable.

> 
> This shot my load up considerably, not sure what was causing it, I
> put everything back (except I increased my cache to 3500 instead of
> 2000 and left the threads at 32) and the load came back down. 
> 
> I can't really figure out where in the background things are having
> problems, I had about 4 slapd processes that were in "R" state chewing
> 95% of the CPU, and wouldn't die when I tried to restart. Memory was
> 95% used, swap was only 5% used.
> 
> Micah
> 

-- 
Kirk Turner-Rustin
Information Systems
Ohio Wesleyan University
http://www.owu.edu
ktrustin@owu.edu