From: Luke Lonergan <llonergan@greenplum.com>
To: peter royal <peter.royal@pobox.com>
To: pgsql-performance@postgresql.org
Subject: Re: help tuning queries on large database
Date: Mon, 09 Jan 2006 13:29:30 -0800
Message-ID: <BFE815BA.1A2F3%llonergan@greenplum.com> (raw)
In-Reply-To: <5CBD78E6-FB37-442C-B7E9-6F1DA03C8AEC@pobox.com>
Peter,
On 1/9/06 12:59 PM, "peter royal" <peter.royal@pobox.com> wrote:
> Overall, I got a 50% boost in the overall speed of my test suite by
> using XFS and the 16k read-ahead.
Yes, it all looks pretty good for your config, though it looks like you
might be adapter limited with the Areca - you should have seen a read time
with XFS of about 17 seconds.
OTOH - with RAID5, you are probably about balanced, you should see a read
time of about 19 seconds and instead you'll get your 22 which isn't too big
of a deal.
> Thanks for the help!
Sure - no problem!
BTW - I'm running tests right now with the 3Ware 9550SX controllers. Two of
them on one machine running simultaneously with 16 drives and we're getting
800MB/s sustained read times. That's a 32GB file read in 40 seconds (!!)
At that rate, we're only about 3x slower than memory access (practically
limited at around 2GB/s even though the system bus peak is 10GB/s). So, the
point is, if you want to get close to your "warm" speed, you need to get
your disk I/O as close to main memory speed as you can. With parallel I/O
you can do that (see Bizgres MPP for more).
- Luke
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Reply to all the recipients using the --to and --cc options:
reply via email
To: pgsql-performance@postgresql.org
Cc: llonergan@greenplum.com, peter.royal@pobox.com
Subject: Re: help tuning queries on large database
In-Reply-To: <BFE815BA.1A2F3%llonergan@greenplum.com>
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
This inbox is served by DDX for PostgreSQL; see mirroring instructions
for how to clone and mirror all data and code used for this inbox