DomainKey-Signature: a=rsa-sha1; c=nofws; d=gmail.com; s=gamma;
	h=message-id:date:from:to:subject:cc:in-reply-to:mime-version:content-type:content-transfer-encoding:content-disposition:references;
	b=w2R3DbUXRykN6tmyXLkNpHpSyQ4iQ95j0KCA3I0ZY7u/pWRmatu1KPRvB6af7O0A+2bg0fc18hiyOq9CiTkr24TZWnYEQe/viHehHJgiCGmSeh1t51WfNwlbFQJT7tG+W5+hs4FeqXbQ6Ai+tnvugeEOMPVyLy6qLrOHbY6GmDY=
Message-ID: <dcc563d10805161318o6b8ce38cn3301ee622bfe3a13@mail.gmail.com>
Date: Fri, 16 May 2008 14:18:02 -0600
From: "Scott Marlowe" <scott.marlowe@gmail.com>
To: "Roberto Edwins" <robertoedwins@gmail.com>
Subject: Re: Setting up of a large database
Cc: pgsql-admin@postgresql.org
In-Reply-To: <97b70d40805161050w1c02e076o89253dc503d352ea@mail.gmail.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=ISO-8859-1
Content-Transfer-Encoding: 7bit
Content-Disposition: inline
References: <97b70d40805161050w1c02e076o89253dc503d352ea@mail.gmail.com>

On Fri, May 16, 2008 at 11:50 AM, Roberto Edwins
<robertoedwins@gmail.com> wrote:
> What are the values that should go in postgresql.conf for a larga
> database, a couple of which tables shall contain over 100 million
> records?

Well, that really kind of depends.  Is this a database to handle lots
of users with small bank style transactions?  Or is it designed to
hold the text type content for millions of users but with only a
handful accessing it at once?  Or maybe it's a reporting database with
lots of statistical data that will be accessed by one or two people at
a time but do huge ugly queries.

It really all kind of depends.

A good idea is to keep from setting max clients any higher than you
have to.  Set shared_buffers around 25% of total memory to start, and
set work_mem to about 8M or so.  Note that it's easy to use up a lot
of memory fast with higher work_mem settings because each sort / hash
agg etc can use up work_mem worth of memory.  100 users times 2 sorts
each would equal 8M*200 or 1600M with a setting of 8M work_mem.

Most importantly, make small incremental changes and benchmark with a
realistic load, it's the only way to be sure.