postgresql benchmarking

[Boris Samorodov]

On Wed, 14 Mar 2007 03:16:10 -0400 Kris Kennaway wrote:
> On Wed, Mar 14, 2007 at 09:42:00AM +0300, Boris Samorodov wrote:
> > On Tue, 13 Mar 2007 20:55:46 -0400 Kris Kennaway wrote:
> > 
> > > On the same test postgresql is indeed faster, I haven't finished
> > > benchmarking it yet though.
> > 
> > Kris, I'm interested at postgresql benchmarking at two-xeon-quad-core
> > server with 3WARE RAID adapter. Can you give me some tips?
> > Thank you.

> Well I'm using sysbench and the collection of patches in my
> kris-contention p4 branch.  Sorry, I don't have an extracted patch set
> available for you to try, but we're hoping to get the most important
> patches into 7.0 within a week or two.

> On the pgsql side, disable the update_process_titles option (or
> whatever it is called), because this has a 33% performance overhead
> for me.  Also if you're testing over localhost, use a unix socket

As for me, I don't like testing with local clients.

> instead of letting it use tcp because the latter is easily saturated
> when you have many CPUs running client threads, and peak performance
> is less than half of what it is over a unix socket.  ULE is faster on
> 7.0, but currently has some performance problems at intermediate
> loads; Jeff hopes to fix this soon.

Got it, thanks!


WBR
-- 
Boris Samorodov (bsam)
Research Engineer, http://www.ipt.ru Telephone & Internet SP
FreeBSD committer, http://www.FreeBSD.org The Power To Serve