Gavin Henry wrote:
> <quote who="Garry">
>> Peter Schober wrote:
>>> * Garry <gkg at gmx.de> [2007-04-10 09:05]:
>>>> on a customer box (with ~250 active users) we've noticed massive
>>>> load on the LDAP server ... this box is pretty well equipped (dual
>>>> Xeon = 4 cores, w/ 6GB memory), so the box performance itself
>>>> shouldn't be much of a problem ... anyway, LDAP numbers are pretty
>>>> disturbing at times ...
>>> note that slapd should hardly make *any* use of your CPUs (unless for
>>> slapindex, which only needs to be run once after changing your indices
>>> in slapd.conf) when properly configured (bdb backend, DB_CONFIG).
>> Well, first results this morning with all the users "storming" the
>> server as they usually do around 8AM resulted in EXTREME improvement in
>> performance ... hardly any lag, system load stays nicely in the 0.5-0.8
>> range, slapd hardly takes up any CPU (47 CPU seconds since restart last
>> night ... normally, it took up 47 CPU seconds in something like 30
>> seconds on the clock or less ...)
>>
>> So, thanks everybody who pointed out the real problem here ;)
> 
> What version? Still 2.2. and ldbm?
> 
> Did you just add better/more indexes then?

Kept the same 2.2 version, but BDB w/ two additional indexes:

index objectClass                       eq,pres
index nisMapName,nisMapEntry            eq,pres,sub

(the other ones mentioned resulted in errors that they were already 
present) as well as

cachesize       5000
checkpoint      512 720

in slapd.conf, and

set_cachesize   0       52428800        0

(among other things) in DB_CONFIG ...


-garry