Date: Sat, 14 May 2005 08:37:22 -0500
From: "Jim C. Nasby" <decibel@decibel.org>
To: Manfred Koizar <mkoi-pg@aon.at>
Cc: Tom Lane <tgl@sss.pgh.pa.us>, Dave Held <dave.held@arrayservicesgrp.com>,
	pgsql-performance@postgresql.org
Subject: Re: Sort and index
Message-ID: <20050514133722.GO49630@decibel.org>
References: <19631.1113966086@sss.pgh.pa.us>
	<20050420034041.GC58835@decibel.org>
	<20050423015404.GV58835@decibel.org>
	<4269.1114222086@sss.pgh.pa.us>
	<20050423030002.GY58835@decibel.org>
	<6536.1114232440@sss.pgh.pa.us>
	<20050424220146.GN58835@decibel.org>
	<ssa481td7d7jpl31rqdf9m0c64iui6qqsf@email.aon.at>
	<20050511211516.GX31103@decibel.org>
	<45678156iqld6sd5n63baovuvqj73palmr@email.aon.at>
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <45678156iqld6sd5n63baovuvqj73palmr@email.aon.at>
User-Agent: Mutt/1.5.8i

On Thu, May 12, 2005 at 08:54:48PM +0200, Manfred Koizar wrote:
> On Wed, 11 May 2005 16:15:16 -0500, "Jim C. Nasby" <decibel@decibel.org>
> wrote:
> >> This is divided by the number of index columns, so the index correlation
> >> is estimated to be 0.219.
> >
> >That seems like a pretty bad assumption to make.
> 
> Any assumption we make without looking at entire index tuples has to be
> bad.  A new GUC variable secondary_correlation introduced by my patch at
> least gives you a chance to manually control the effects of additional
> index columns.

It seems it would be much better to gather statistics on any
multi-column indexes, but I know that's probably beyond what's
reasonable for your patch.

Also, my data (http://stats.distributed.net/~decibel) indicates that
max_io isn't high enough. Look specifically at timing2.log compared to
timing.log. Thouggh, it is possibile that this is because of having
random_page_cost set to 1.1 (if I set it much higher I can't force the
index scan because the index estimate actually exceeds the cost of the
seqscan with the disable cost added in).

> >It depends on the patches, since this is a production machine. Currently
> >it's running 7.4.*mumble*,
> 
> The patch referenced in
> http://archives.postgresql.org/pgsql-hackers/2003-08/msg00931.php is
> still available.  It doesn't touch too many places and should be easy to
> review.  I'm using it and its predecessors in production for more than
> two years.  Let me know, if the 74b1 version does not apply cleanly to
> your source tree.

Looks reasonable; I'll give it a shot on 8.0 once I have replication
happening.
-- 
Jim C. Nasby, Database Consultant               decibel@decibel.org 
Give your computer some brain candy! www.distributed.net Team #1828

Windows: "Where do you want to go today?"
Linux: "Where do you want to go tomorrow?"
FreeBSD: "Are you guys coming, or what?"