DomainKey-Signature: a=rsa-sha1; q=dns; c=nofws; s=dk20050327;
 d=earthlink.net;
	b=l0yqXtCyDQ/KllgHFCLukR6f3L6Asf+6etDH1QWPKcIcCdJRfA0VHBnuKqq5A5LU;
	h=Received:Message-Id:X-Mailer:Date:To:From:Subject:In-Reply-To:References:Mime-Version:Content-Type:X-ELNK-Trace:X-Originating-IP;
Message-Id: <7.0.1.0.2.20060225063934.00adfec0@earthlink.net>
Date: Sat, 25 Feb 2006 06:56:33 -0500
To: "Luke Lonergan" <llonergan@greenplum.com>,
 pgsql-performance@postgresql.org
From: Ron <rjpeace@earthlink.net>
Subject: Re: Reliability recommendations
In-Reply-To: <C02537B9.1DC8D%llonergan@greenplum.com>
References: <43FFF4EF.6060601@paradise.net.nz>
	<C02537B9.1DC8D%llonergan@greenplum.com>
Mime-Version: 1.0
Content-Type: text/plain; charset="us-ascii"; format=flowed

At 01:22 AM 2/25/2006, Luke Lonergan wrote:
>Mark,
>
>On 2/24/06 10:10 PM, "Mark Kirkwood" <markir@paradise.net.nz> wrote:
>
> > Well, since this is always fun (2G memory, 3Ware 7506, 4xPATA), writing:
> >
> > $ dd if=/dev/zero of=/data0/dump/bigfile bs=8k count=500000
> > 500000 records in
> > 500000 records out
> > 4096000000 bytes transferred in 32.619208 secs (125570185 bytes/sec)
>
> > Reading:
> >
> > $ dd of=/dev/null if=/data0/dump/bigfile bs=8k count=500000
> > 500000 records in
> > 500000 records out
> > 4096000000 bytes transferred in 24.067298 secs (170189442 bytes/sec)
>
>Not bad at all!  I have one of these cards in my home machine running WinXP
>and it's not nearly this fast.
>
> > Hmmm - a bit humbled by Luke's machinery :-), however, mine is probably
> > competitive on (MB/s)/$....
>
>Not sure - the machines I cite are about $10K each.  The machine you tested
>was probably about $1500 a few years ago (my guess), and with a 5:1 ratio in
>speed versus about a 6:1 ratio in price, we're not too far off in MB/s/$
>after all :-)
>
> > It would be interesting to see what Dan's system would do on a purely
> > sequential workload - as 40-50MB of purely random IO is high.
>
>Yeah - that is really high if the I/O is really random.  I'd normally expect
>maybe 500-600 iops / second and if each IO is 8KB, that would be 4MB/s.  The
>I/O is probably not really completely random, or it's random over cachable
>bits of the occupied disk area.
Side note: the new WD 150GB Raptors (10Krpm 1.5Gbps SATA w/ NCQ 
support) have benched at ~1000 IOps _per drive_

http://www.storagereview.com/articles/200601/WD1500ADFD_4.html

(Now if we can just get WD to make a 300GB Raptor, increase that 
wimpy 16MB buffer, and implement 6Gbps SATA...;-) )

An array of these things plugged into a PCI-E <-> SATA RAID 
controller with 1-2GB of cache should set a new bar for performance 
as well as making that performance more resilient than ever to 
variations in usage patterns.