Sébastien Georget wrote:
I did not mention it because I was looking for a general way to find what's the bottleneck in an openldap installation. I can stress the server and play with the configuration to find what parameters improve performances (reduce failed authentications) but it's a long and not so easy process. I thought that server-side performances mesuring tools would have help in tuning the configuration with information such as
It sounds like you're asking a pretty general question about code profiling then, not something specific to LDAP or OpenLDAP.
- time spent processing filters - time spent processing acl - time spent in backend specific code
On x86 Linux valgrind can be useful here (with calltree module). On Solaris and Linux I use FunctionCheck 1.5.4 for profiling. (Originally written by Yannick Perret, significantly enhanced by me.)
http://freshmeat.net/projects/fncchk/
- number of connections rejected - peak request/s - ... Here is the configuration I am using : - openldap 2.2.27 with 8 bdb backends (1 + 7 subordinates) - 250 entries, will certainly grow to 2000-3000 > What sort of entry cache do you have? cachesize 2000 default for other parameters I plan to make tests with : idlcachesize, conn_max_pending, timelimit, idletimeout Since the bdb files are very small (<200k) I supposed that they stay in memory and I dit not look at DB_CONFIG files, should I ?
Probably. Use db_stat -m, that will tell you whether the current (default) settings are working well or not.
-- -- Howard Chu Chief Architect, Symas Corp. Director, Highland Sun http://www.symas.com http://highlandsun.com/hyc Symas: Premier OpenSource Development and Support
