Message-ID: <4BA83F99.5020607@2ndquadrant.com>
Date: Tue, 23 Mar 2010 00:12:09 -0400
From: Greg Smith <greg@2ndquadrant.com>
User-Agent: Thunderbird 2.0.0.24 (X11/20100317)
MIME-Version: 1.0
To: Carlo Stonebanks <stonec.register@sympatico.ca>
CC: pgsql-performance@postgresql.org
Subject: Re: Got that new server, now it's time for config!
References: <ho8r9s$2i4p$1@news.hub.org>
In-Reply-To: <ho8r9s$2i4p$1@news.hub.org>
Content-Type: text/plain; charset=ISO-8859-1; format=flowed
Content-Transfer-Encoding: 7bit

Carlo Stonebanks wrote:
> So, we have the hardware, we have the O/S - but I think our config 
> leaves much to be desired. Typically, our planner makes nad decisions, 
> picking seq scan over index scan, where index scan has a better result.
>

You're not setting effective_cache_size, so I wouldn't expect it to ever 
choose an index scan given the size of your data set.  The planner 
thinks that anything bigger than 128MB isn't likely to fit in RAM by 
default, which favors sequential scans.  That parameter should probably 
be >24GB on your server, so it's off by more than two orders of magnitude.

> wal_sync_method = open_sync

This is a scary setting to be playing with on Linux when using ext3 
filesystems due to general kernel bugginess in this area.  See 
http://archives.postgresql.org/pgsql-hackers/2007-10/msg01310.php for an 
example.  I wouldn't change this from the default in your position if 
using that filesystem.

I'd drastically increase effective_cache_size, put wal_sync_method back 
to the default, and then see how things go for a bit before tweaking 
anything else.  Nothing else jumped out as bad in your configuration 
besides the extremely high logging levels, haven't looked at it that 
carefully yet though.

-- 
Greg Smith  2ndQuadrant US  Baltimore, MD
PostgreSQL Training, Services and Support
greg@2ndQuadrant.com   www.2ndQuadrant.us