Date: Tue, 11 Sep 2007 15:32:42 -0500
From: Decibel! <decibel@decibel.org>
To: Carlo Stonebanks <stonec.register@sympatico.ca>
Cc: 'Scott Marlowe' <scott.marlowe@gmail.com>,
	'Alvaro Herrera' <alvherre@commandprompt.com>,
	pgsql-performance@postgresql.org
Subject: Re: Performance on 8CPU's and 32GB of RAM
Message-ID: <20070911203242.GY38801@decibel.org>
References: <fbkn5d$cfq$1@news.hub.org>
 <20070904230639.GA23731@alvh.no-ip.org>
	<dcc563d10709041615s305ae98bm92dee61ab199b7c4@mail.gmail.com>
	<3DCCF196F82444998B33F40C97B71CA4@serenity>
MIME-Version: 1.0
Content-Type: multipart/signed; micalg=pgp-sha1;
	protocol="application/pgp-signature"; boundary="MMGYF6kLVCwQWRVm"
Content-Disposition: inline
In-Reply-To: <3DCCF196F82444998B33F40C97B71CA4@serenity>
User-Agent: Mutt/1.5.13 (2006-08-11)


--MMGYF6kLVCwQWRVm
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
Content-Transfer-Encoding: quoted-printable

On Wed, Sep 05, 2007 at 11:06:03AM -0400, Carlo Stonebanks wrote:
> Unfortunately, LINUX is not an option at this time. We looked into it; th=
ere
> is no *NIX expertise in the enterprise. However, I have raised this issue=
 in
> various forums before, and when pressed no one was willing to say that "*=
NIX
> *DEFINITELY* outperforms Windows" for what my client is doing (or if it d=
id
> outperform Windows, that it would outperform so significantly that it
> merited the move).
>=20
> Was this incorrect? Can my client DEFINITELY expect a significant
> improvement in performance for what he is doing?

Since we don't know your actual workload, there's no way to predict
this. That's what benchmarking is for. If you haven't already bought the
hardware, I'd strongly recommend benchmarking this before buying
anything, so that you have a better idea of what your workload looks
like. Is it I/O-bound? CPU-bound? Memory?

One of the fastest ways to non-performance in PostgreSQL is not
vacuuming frequently enough. Vacuum more, not less, and control IO
impact via vacuum_cost_delay. Make sure the FSM is big enough, too.

Unless your database is small enough to fit in-memory, your IO subsystem
is almost certainly going to kill you. Even if it does fit in memory, if
you're doing much writing at all you're going to be in big trouble.
--=20
Decibel!, aka Jim Nasby                        decibel@decibel.org
EnterpriseDB      http://enterprisedb.com      512.569.9461 (cell)

--MMGYF6kLVCwQWRVm
Content-Type: application/pgp-signature
Content-Disposition: inline

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2.0.3 (FreeBSD)

iD8DBQFG5vtqdO30qud8SkgRAkNLAJ9Ct/oHftNOKvTbSJsyYzXABEhFKACfZvJ/
HZxf/IRFxCsq6ez7cnm/JUw=
=mAQ4
-----END PGP SIGNATURE-----

--MMGYF6kLVCwQWRVm--