Re: strange performance problem

[xoror@infuse.org]

On Sun, 23 Mar 2003, [ISO-8859-1] Michael Ströder wrote:

> Cuong bui wrote:
> > 
> > For this purpose i've written an multithreaded application to simulate 
> > simultaneous access to the ldap server. This application creates an variable 
> > number of threads that simulate an client.
> 
> Another wild guess: How did you implement this multi-threaded LDAP 
> application? Did you make sure you don't have any thread-safety issues? I'd 
> suggest to use forking processes opening separate LDAP connections instead.
> 

As far as i can see it is thread safe. i've created N number of workers
who all have their own variables. Access to the task queue is done with a
mutex. Each thread has it's own connection. 

surely if it were thread safety issues then it would manifest
more randomly.  Not consequent after 4 test runs (and all
consequent testruns suffer from it, even ldapsearch utility
suffer from this point on). And when you restore die bdb files,
all is fine again. 

i'm poking around with the db_stat and other db_* utilities to see if it's
maybe a lock problem. thanks for the suggestion anyway.

I've found some documentation on the conflict matrix. The described model
at http://www.sleepycat.com/docs/ref/lock/stdmode.html
differs from what i'm seeing (output of db_stat).

they have an 6x6 matrix where i have 9x9 matrix.