Re: [Fwd: Re: (ITS#4108) LWL causing core dumps in HEAD]

[Quanah Gibson-Mount <quanah@symas.com>]

--On Sunday, October 30, 2005 1:46 AM -0800 Howard Chu <hyc@symas.com> 
wrote:

> I note that the code in HEAD now is noticably slower than RE23 or RE22,
> and far less consistent.
>
> Using the slapd-bind program against RE22 or RE23 I can consistently get
> 10000 binds across a 100Mbit ethernet in 7-8 seconds. On localhost it
> only takes 2.6 seconds.
>
> Using HEAD, localhost takes 3-5 seconds. Across 100Mbit ethernet the time
> varies from as low as 10 seconds to as much as 45 seconds. (And there are
> only 3 machines on this network, no other activity.)
>
> Some of the slowdowin is probably due to the double pool_submit that
> occurs for read events...

Some initial benchmarking I did last night of HEAD with LWL in place found 
it to be nearly 2,000 searches/second slower than 2.3, which is a fairly 
substantial hit.  Head ran at 6,153.8 searches/second, 2.3 at 7,975.396 
searches/second.  Another way to think of it is that HEAD with LWL is 23% 
slower than 2.3.

--Quanah

--
Quanah Gibson-Mount
Product Engineer
Symas Corporation
Packaged, certified, and supported LDAP solutions powered by OpenLDAP:
<http://www.symas.com>