Date: Fri, 1 Dec 2006 02:15:54 +0100
From: Tobias Brox <tobias@nordicbet.com>
To: nospam@hardgeus.com
Cc: Paul Lathrop <plathrop@squaretrade.com>, pgsql-performance@postgresql.org
Subject: Re: Defining performance.
Message-ID: <20061201011554.GA27123@oppetid.no>
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
Organization: Group Nordicbet
User-Agent: Mutt/1.5.9i

[nospam@hardgeus.com - Thu at 06:37:12PM -0600]
> As my dataset has gotten larger I have had to throw more metal at the
> problem, but I have also had to rethink my table and query design.  Just
> because your data set grows linearly does NOT mean that the performance of
> your query is guaranteed to grow linearly!  A sloppy query that runs OK
> with 3000 rows in your table may choke horribly when you hit 50000.

Then some limit is hit ... either the memory cache, or that the planner
is doing an unlucky change of strategy when hitting 50000.

Anyway, it's very important when testing queries that they actually are
tested on a (copy of the) production database, and not on an empty
database or a database containing some random test data.  If testing
queries off the production database, it's important to have equal
hardware and configuration on both servers.