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

Re: slapd lightweight dispatcher

Pierangelo Masarati wrote:
Your numbers are quite interesting. About the workload tool issue, we
could do some work to improve the existing test suite in tests/progs. The
current versions in HEAD look much more flexible than ever: now they allow
random search/bind on entries looked up from the database, we're using
them to benchmork stock and custom slapd on different architectures. Maybe a more consistent and coordinated development would turn them into
something more valuable?.

Yes, that sounds like a good idea. Right now it's as if we're using OpenLDAP slapd to benchmark SLAMD/java, not the other way around. We need a real benchmarking tool that doesn't waste half its CPU cycles getting out of its own way. With a slightly modified slapd-bind.c (to duplicate the search/bind pattern of the SLAMD job) I can get around 25 times as many authentications/second as a single SLAMD client (both running on the same machine).

As an aside, kudos to Sun's Niagara team. This T2000 runs OpenLDAP at "ludicrous speed." Too bad we don't have a dedicated one for running the clients.

 -- Howard Chu
 Chief Architect, Symas Corp.  http://www.symas.com
 Director, Highland Sun        http://highlandsun.com/hyc
 OpenLDAP Core Team            http://www.openldap.org/project/