To: Paul Lathrop <plathrop@squaretrade.com>
cc: pgsql-performance@postgresql.org
Subject: Re: Defining performance. 
In-reply-to: <456F624F.2090906@squaretrade.com> 
References: <456F624F.2090906@squaretrade.com>
Comments: In-reply-to Paul Lathrop <plathrop@squaretrade.com>
	message dated "Thu, 30 Nov 2006 14:59:27 -0800"
Date: Thu, 30 Nov 2006 19:26:36 -0500
Message-ID: <8312.1164932796@sss.pgh.pa.us>
From: Tom Lane <tgl@sss.pgh.pa.us>

Paul Lathrop <plathrop@squaretrade.com> writes:
> ... When I joined the company last year, the databases were
> deployed on 12-disk RAID5 arrays on dual-proc AMD machines with 4Gb of
> RAM, running Debian Woody and Postgres 7.2. These systems seemed to
> suffer a gradually decreasing performance accompanied by a gradually
> growing disk space usage. The DBA had come to the conclusion that the
> VACUUM command did/does not work on these systems, because even after a
> VACUUM FULL, the size of the database was continually increasing.

The very first thing you need to do is get off 7.2.

After that, I'd recommend looking at *not* using VACUUM FULL.  FULL is
actually counterproductive in a lot of scenarios, because it shrinks the
tables at the price of bloating the indexes.  And 7.2's poor ability to
reuse index space turns that into a double whammy.  Have you checked
into the relative sizes of tables and indexes and tracked the trend over
time?

			regards, tom lane