From: Gregory Stark <stark@enterprisedb.com>
To: "PFC" <lists@peufeu.com>
Cc: <pgsql-performance@postgresql.org>
Subject: Re: Postgres Benchmark Results
In-Reply-To: <op.tsmrv71ecigqcu@apollo13> (lists@peufeu.com's message of "Sun,
	20 May 2007 16:58:45 +0200")
Organization: EnterpriseDB
References: <op.tsmrv71ecigqcu@apollo13>
Date: Tue, 22 May 2007 09:16:56 +0100
Message-ID: <87hcq5l1c7.fsf@oxford.xeocode.com>
User-Agent: Gnus/5.110006 (No Gnus v0.6) Emacs/21.4 (gnu/linux)
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii


What's interesting here is that on a couple metrics the green curve is
actually *better* until it takes that nosedive at 500 MB. Obviously it's not
better on average hits/s, the most obvious metric. But on deviation and
worst-case hits/s it's actually doing better.

Note that while the average hits/s between 100 and 500 is over 600 tps for
Postgres there is a consistent smattering of plot points spread all the way
down to 200 tps, well below the 400-500 tps that MySQL is getting.

Some of those are undoubtedly caused by things like checkpoints and vacuum
runs. Hopefully the improvements that are already in the pipeline will reduce
them.

I mention this only to try to move some of the focus from the average
performance to trying to remove the pitfalls that affact 1-10% of transactions
and screw the worst-case performance. In practical terms it's the worst-case
that governs perceptions, not average case.

-- 
  Gregory Stark
  EnterpriseDB          http://www.enterprisedb.com