[Date Prev][Date Next] [Chronological] [Thread] [Top]

Re: Connections timing out to OpenLDAP server





--On Friday, August 19, 2005 4:16 PM -0500 Ben Beuchler <insyte@gmail.com> wrote:

On 8/19/05, Quanah Gibson-Mount <quanah@stanford.edu> wrote:

> Ah.  The only thing I'm setting from those options is several "index"
> entries.  Anything in particular you recommend I experiment with?

cachesize and idlcachesize

We only have ~1300 entries in our database, so I left cachesize at the default (1000) and set idlcachesize to 200. It'll be interesting to see how that changes things.

I would set up a cachesize then that allows you to cache all your entries, plus room for growth.


Note that the general rule of thumb for idlcachesize is 3*cachesize (so in your case, around 4500, rather than 200).


That does seem small, but after reading the documentation you linked
to and playing with db_stat, it appears we were getting 85% to 97%
cache hits, depending on the attribute.

In the name of experimentation, I doubled it to 4MB.  I'm going to
spend some time reading about the bdb tweaks and see what I can
optimize.  However, since our database is so very small, I'm not
expecting dramatic changes...

Sounds good. :)

Another way to test this (on a development server) is via slapadd. When slapadd runs out of cache space, it slows down significantly. So if it takes you a longer time to slapadd with 2MB than it does with 4MB, you definitely had too small a cache. And you could always see if there is any difference between say 4MB and 8MB, just for kicks. ;)

--Quanah

--
Quanah Gibson-Mount
Principal Software Developer
ITSS/Shared Services
Stanford University
GnuPG Public Key: http://www.stanford.edu/~quanah/pgp.html