Message-ID: <450FEC1E.7080600@logix-tt.com>
Date: Tue, 19 Sep 2006 15:09:50 +0200
From: Markus Schaber <schabi@logix-tt.com>
Organization: Logical Tracking and Tracing International AG, Switzerland
User-Agent: Thunderbird 1.5.0.5 (X11/20060812)
MIME-Version: 1.0
To: pgsql-performance@postgresql.org
Subject: Re: High CPU Load
References: <1158239339.5226.17.camel@localhost.localdomain>	
	<450A75DE.6050102@logix-tt.com>	
	<1158590646.5665.71.camel@localhost.localdomain>	
	<450FBE15.9010905@logix-tt.com>
	<1158670101.5204.14.camel@localhost.localdomain>
In-Reply-To: <1158670101.5204.14.camel@localhost.localdomain>
Content-Type: text/plain; charset=ISO-8859-15
Content-Transfer-Encoding: quoted-printable

Hi, Jerome,

J=E9r=F4me BENOIS wrote:

>>> Now i Have 335 concurrent connections, i decreased work_mem parameter=
 to
>>> 32768 and disabled Hyper Threading in BIOS. But my CPU load is still
>>> very important.
>> What are your settings for commit_siblings and commit_delay?
> It default :
>=20
> #commit_delay =3D 01   # range 0-100000, inmicroseconds
> #commit_siblings =3D 5 # range 1-1000

You should uncomment them, and play with different settings. I'd try a
commit_delay of 100, and commit_siblings of 5 to start with.

> I plan to return to previous version : 7.4.6 in and i will reinstall al=
l
> in a dedicated server in order to reproduce and solve the problem.

You should use at least 7.4.13 as it fixes some critical buts that were
in 7.4.6. They use the same on-disk format and query planner logic, so
they should not have any difference.

I don't have much more ideas what the problem could be.

Can you try to do some profiling (e. G. with statement logging) to see
what specific statements are the one that cause high cpu load?

Are there other differences (besides the PostgreSQL version) between the
two installations? (Kernel, libraries, other software...)

HTH,
Markus
--=20
Markus Schaber | Logical Tracking&Tracing International AG
Dipl. Inf.     | Software Development GIS

Fight against software patents in Europe! www.ffii.org
www.nosoftwarepatents.org