Date: Tue, 5 Jun 2007 10:42:36 -0400 (EDT)
From: Greg Smith <gsmith@gregsmith.com>
To: pgsql-performance@postgresql.org
Subject: Re: dbt2 NOTPM numbers
In-Reply-To: <4665703E.3040905@bluegap.ch>
Message-ID: <Pine.GSO.4.64.0706051028240.28129@westnet.com>
References: <466428F7.5010808@bluegap.ch> <46642B33.6030101@enterprisedb.com>
	<46642F2E.3080509@bluegap.ch> <4664748B.2060304@enterprisedb.com>
	<4665703E.3040905@bluegap.ch>
MIME-Version: 1.0
Content-Type: TEXT/PLAIN; charset=US-ASCII; format=flowed

On Tue, 5 Jun 2007, Markus Schiltknecht wrote:

> I'm really wondering, if the RAID 6 of the ARECA 1260 hurts so badly

All of your disk performance tests look reasonable; certainly not slow 
enough to cause the issue you're seeing.  The only thing I've seen in this 
thread that makes me slightly suspicious is:

> Oh, something else that's probably worth thinking about (and just came 
> to my mind again): the XFS is on a lvm2, on that RAID6.

There have been reports here from reliable sources that lvm can introduce 
performance issues; 
http://archives.postgresql.org/pgsql-performance/2006-07/msg00276.php is 
one example.  Your dd test results suggest you're getting decent thoughput 
though, so you don't seem to be suffering too much from that.

Anyway, did you post your postgresql.conf yet?  I don't remember seeing 
it.  From looking at your latest data, my first guess would be there's 
something wrong there that's either causing a) the system to be 
checkpointing constantly, or b) not enough memory allocated to allow 
queries to execute properly.  The fact that all your long statement times 
come from SELECTs suggests to me that playing with the WAL parameters like 
fsync isn't likely to help here.

--
* Greg Smith gsmith@gregsmith.com http://www.gregsmith.com Baltimore, MD