Date: Tue, 8 May 2007 16:48:34 +0200
From: "Steinar H. Gunderson" <sgunderson@bigfoot.com>
To: Susan Russo <russo@morgan.harvard.edu>
Cc: pgsql-performance@postgresql.org, harvsys@morgan.harvard.edu
Subject: Re: specific query (not all) on Pg8 MUCH slower than Pg7
Message-ID: <20070508144834.GA5678@uio.no>
Mail-Followup-To: Susan Russo <russo@morgan.harvard.edu>,
	pgsql-performance@postgresql.org, harvsys@morgan.harvard.edu
References: <200705081418.l48EIY8V026360@larrys.harvard.edu>
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Disposition: inline
In-Reply-To: <200705081418.l48EIY8V026360@larrys.harvard.edu>
User-Agent: Mutt/1.5.13 (2006-08-11)

On Tue, May 08, 2007 at 10:18:34AM -0400, Susan Russo wrote:
> explain analyze output on Pg7.3.2:
> 
>                ->  Index Scan using dbxref_idx2 on dbxref dx  (cost=0.00..5.83 rows=1 width=21) (actual time=25.58..25.58 rows=0 loops=1)
>                      Index Cond: ((accession >= 'AY851043'::character varying) AND (accession < 'AY851044'::character varying))
>                      Filter: (accession ~~ 'AY851043%'::text)
> 
> explain analyze output on Pg8.1.4:
> 
>                ->  Seq Scan on dbxref dx  (cost=0.00..47923.91 rows=1 width=21) (actual time=2463.646..2463.646 rows=0 loops=1)
>                      Filter: ((accession)::text ~~ 'AY851043%'::text)

This is almost all of your cost. Did you perchance initdb the 8.1.4 cluster
in a non-C locale? You could always try

  CREATE INDEX test_index ON dbxref (accession varchar_pattern_ops);

which would create an index that might be more useful for your LIKE query,
even in a non-C locale.

/* Steinar */
-- 
Homepage: http://www.sesse.net/