Date: Fri, 19 May 2006 15:21:35 -0500
From: "Jim C. Nasby" <jnasby@pervasive.com>
To: Jean-Paul Argudo <jean-paul@argudo.org>
Cc: Olivier Andreotti <olivier.andreotti@gmail.com>,
	pgsql-performance@postgresql.org
Subject: Re: Benchmarck PostgreSQL 8.1.4 MySQL 5.0.20 and Oracle
Message-ID: <20060519202134.GM64371@pervasive.com>
References: <d5d40b330605180257w2a0cb2dei53d81cbe0d7405e@mail.gmail.com>
	<446C510A.9040609@argudo.org>
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <446C510A.9040609@argudo.org>
User-Agent: Mutt/1.5.11

On Thu, May 18, 2006 at 12:48:42PM +0200, Jean-Paul Argudo wrote:
> > autovaccuum = on
> 
> Thats a critic point. Personaly I dont use autovacuum. Because I just
> don't want a vacuum to be started ... when the server is loaded :)
> 
> I prefer control vacuum process, when its possible (if its not,
> autovacuum is the best choice!), for example, a nighlty vacuum...

This can be problematic for a benchmark, which often will create dead
tuples at a pretty good clip.

In any case, if you are going to use autovacuum, you should cut all the
thresholds and scale factors in half, and set cost_delay to something (I
find 5-10 is usually good).

Depending on your write load, you might need to make the bgwriter more
aggressive, too.

If you can graph some metric from your benchmark over time it should be
pretty easy to spot if the bgwriter is keeping up with things or not; if
it's not, you'll see big spikes every time there's a checkpoint.

> A question for you: after setting up your test database, did you launch
> a vacuum full analyze of it ?

Why would you vacuum a newly loaded database that has no dead tuples?
-- 
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