Date: Fri, 23 Feb 2007 11:43:56 -0600
From: "Jim C. Nasby" <jim@nasby.net>
To: Richard Huxton <dev@archonet.com>
Cc: "Campbell, Lance" <lance@uiuc.edu>, pgsql-performance@postgresql.org
Subject: Re: Recommended Initial Settings
Message-ID: <20070223174356.GB19527@nasby.net>
References: <A3AC4FA47DC0B1458C3E5396E685E63302395CE7@SAB-DC1.sab.uiuc.edu>
	<45DF1660.101@archonet.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <45DF1660.101@archonet.com>
User-Agent: Mutt/1.5.13 (2006-08-11)

If you're doing much updating at all you'll also want to bump up
checkpoint_segments. I like setting checkpoint_warning just a bit under
checkpoint_timeout as a way to monitor how often you're checkpointing
due to running out of segments.

With a large shared_buffers you'll likely need to make the bgwriter more
aggressive as well (increase the max_pages numbers), though how
important that is depends on how much updating you're doing. If you see
periodic spikes in IO corresponding to checkpoints, that's an indication
bgwriter isn't doing a good enough job.

If everything ends up in memory, it might be good to decrease
random_page_cost to 1 or something close to it; though the database
should just rely on effective_cache to figure out that everything's in
memory.

If you're on pre-8.2, you'll want to cut all the autovacuum parameters
in half, if you're using it.
-- 
Jim Nasby                                            jim@nasby.net
EnterpriseDB      http://enterprisedb.com      512.569.9461 (cell)