Date: Tue, 28 Mar 2006 11:30:47 -0600
From: "Jim C. Nasby" <jnasby@pervasive.com>
To: pgsql-performance@postgresql.org
Subject: Re: simple join uses indexes, very slow
Message-ID: <20060328173047.GN75181@pervasive.com>
References: <20060327134733.72422d23.gry@ll.mit.edu>
	<00c901c65283$401f14d0$8300a8c0@tridecap.com>
	<20060328162908.GB26539@uio.no>
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20060328162908.GB26539@uio.no>
User-Agent: Mutt/1.5.11

On Tue, Mar 28, 2006 at 06:29:08PM +0200, Steinar H. Gunderson wrote:
> On Tue, Mar 28, 2006 at 10:18:25AM -0600, Dave Dutcher wrote:
> >>     "parameters_idx" btree (run, opset_num, step_num, opset,
> > opset_ver,
> >> step, step_ver, name, split, wafers)
> >>     "parameters_opset_idx" btree (opset, step, name)
> >>     "parameters_step_idx" btree (step, name)
> > Have you tried creating some different indexes on parameters?  I don't
> > know if it should matter or not, but I would try some indexes like:
> > 
> > (run, opset_num) //Without all the other columns
> > (opset_num, run) //Backwards
> > (opset_num)
> 
> An index on (A,B,C) can be used for a query on (A,B) or (A), so it doesn't
> really matter. It isn't usable for a query on (B), (C) or (B,C), though. (The
> index rows will get bigger, of course, so you'll need more I/O if you want to
> scan large parts of it, but I guess that's beside the point.)

Note that given how statistics currenly work, there are many situations
where the planner will refuse to use a multi-column index. This probably
won't change until there's some concept of multi-column statistics, at
least for multi-column indexes.
-- 
Jim C. Nasby, Sr. Engineering Consultant      jnasby@pervasive.com
Pervasive Software      http://pervasive.com    work: 512-231-6117
vcard: http://jim.nasby.net/pervasive.vcf       cell: 512-569-9461