Re: syncrepl w/ 2.3 and bdb 4.3 working great

[Howard Chu <hyc@symas.com>]

Dusty Doris wrote:
> > The problems I had with BDB 4.3 (21 & 27) were related to data 
> > loading via slapadd (and the use of IN-MEMORY logs).  The "-q" 
> > function in 2.3 removes the necessity of the IN-MEMORY logs.  The 
> > other issues seen in 21 & 27 were reported by other people, and may 
> > well have been resolved in 4.3.28.
>
> I did run into some snags testing openldap 2.3.7 w/ syncprov.c from HEAD
> and BDB 4.2 or 4.3 today when using hdb as the backend.
>
> When I initially start up the slapd, it starts no problem.  But then
> something happens during the initial search that takes some time.  If 
> I do
> an initial single search and wait for the response, it takes about 10
> seconds and then each search after that point is immediate.
>
> The problem is if I start a bunch of searches when the initial load up
> happens.  In my case the searches are generated from a radius server
> and they are aborted after a few seconds.  The slapd process slowly 
> creeps
> up over 100% when running top and I get a bunch of these in the logs

Funny you should mention this. Quanah and I just spent the last couple 
days profiling / benchmarking back-hdb. While hdb's write performance is 
better than bdb's, its search performance was terrible. Profiling 
identified a couple of bottlenecks that I have now rewritten, and 
back-hdb search performance is now as fast as or faster than back-bdb 
(in CVS HEAD - over a 4x speedup vs 2.3.7). We still have more 
benchmarks to run, but you might want to test the current CVS HEAD and 
see how it behaves with your data.

--
 -- 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/