Date: Thu, 1 Sep 2005 23:01:58 +0100
From: Matthew Sackman <matthew@lshift.net>
To: pgsql-performance@postgresql.org
Subject: Re: Massive performance issues
Message-ID: <20050901220158.GF7131@pongo.lshift.net>
Mail-Followup-To: pgsql-performance@postgresql.org
References: <20050901174231.GE23970@pongo.lshift.net>
	<43176A95.70506@tweakers.net>
	<20050901211359.GD7131@pongo.lshift.net>
	<20050901215245.GA4242@uio.no>
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20050901215245.GA4242@uio.no>
User-Agent: Mutt/1.5.10i

On Thu, Sep 01, 2005 at 11:52:45PM +0200, Steinar H. Gunderson wrote:
> On Thu, Sep 01, 2005 at 10:13:59PM +0100, Matthew Sackman wrote:
> > Well that's the thing - on the queries where it decides to use the index
> > it only reads at around 3MB/s and the CPU is maxed out, whereas when it
> > doesn't use the index, the disk is being read at 60MB/s. So when it
> > decides to use an index, I don't seem to be IO bound at all. Or at least
> > that's the way it seems to me.
> 
> You are I/O bound; your disk is doing lots and lots of seeks. The SATA
> interface is not the bottleneck; the disk's ability to rotate and move its
> heads is.

Ahh of course (/me hits head against wall). Because I've /seen/ it read
at 60MB/s I was assuming that if it wasn't reading that fast then I'm
not IO bound but of course, it's not reading sequentially. That all
makes sense. Been a long day etc... ;-)

Matthew