To: "Mikael Carneholm" <Mikael.Carneholm@WirelessCar.com>
cc: "Olivier Andreotti" <olivier.andreotti@gmail.com>,
	pgsql-performance@postgresql.org
Subject: Re: Benchmarck PostgreSQL 8.1.4 MySQL 5.0.20 and Oracle 10g2 
In-reply-to: <7F10D26ECFA1FB458B89C5B4B0D72C2B459FFB@sesrv12.wirelesscar.com> 
References: <7F10D26ECFA1FB458B89C5B4B0D72C2B459FFB@sesrv12.wirelesscar.com>
Comments: In-reply-to "Mikael Carneholm" <Mikael.Carneholm@WirelessCar.com>
	message dated "Thu, 18 May 2006 16:11:51 +0200"
Date: Thu, 18 May 2006 10:48:20 -0400
Message-ID: <13688.1147963700@sss.pgh.pa.us>
From: Tom Lane <tgl@sss.pgh.pa.us>

"Mikael Carneholm" <Mikael.Carneholm@WirelessCar.com> writes:
> Btw, check you logfile for hints regarding increasing max_fsm_pages, and
> consider increasing checkpoint_segments as well. You could also play
> with more aggressive bgwriter_* params to reduce the risk for long
> vacuum pauses.

Yeah, checkpoint_segments is a really critical number for any
write-intensive situation.  Pushing it up to 30 or more can make a big
difference.  You might want to set checkpoint_warning to a large value
(300 or so) so you can see in the log how often checkpoints are
happening.  You really don't want checkpoints to happen more than about
once every five minutes, because not only does the checkpoint itself
cost a lot of I/O, but there is a subsequent penalty of increased WAL
traffic due to fresh page images getting dumped into WAL.

			regards, tom lane