Re: CPU 0.1% IOWAIT 99% for decisonnal queries

Great !

I'm not an expert but as far as I know, my 15 databases are spread over 4 SCSI RAID disks 73 GB 10K RPM mounted under ext3 mode.
I remember that they where provided by DELL under RAID5 and I asked my system engineer for switching them to standard SCSI because I don't care about security but only about speed and capacity ( maybe this switch was not set properly at this time...).

Thank you for these interesting links: I 've sent them to my system engineer with my two hands !

Amicalement

Patrick
----- Original Message -----
From: Gustavo Franklin Nóbrega - Planae
To: Patrick Vedrines
Cc: performance pgsql
Sent: Thursday, March 24, 2005 12:52 PM
Subject: Re: [PERFORM] CPU 0.1% IOWAIT 99% for decisonnal queries

Good day Patrick!

I can help you to design you disk layout for better perform and security. Please, tell me how many disks (and some specs, like capacity and RPM).

If you want to know more, there is a very interesting article abou benckmark filesystem ( http://linuxgazette.net/102/piszcz.html ). In this article, ReiserFS 3.6, JFS and XFS are in the same level at top depending your application, and ext3 is more slow than others. I believe that version 4 of the ReiserFS is better that version 3.6, but I could not still test it.

Raid0 (striping, more at http://www.pcguide.com/ref/hdd/perf/raid/levels/singleLevel0-c.html) or Raid 0+1 (stripping + mirror, more at http://www.pcguide.com/ref/hdd/perf/raid/levels/multLevel01-c.html) is very insteresting to postgresql. Raid0 provides only performance, and Raid 0+1 provides performance and security. Take a look at this articles and think about to use Raid (http://www.pcguide.com/ref/hdd/perf/raid/)

I'm glad to help. Best regards!

Atenciosamente,

Gustavo Franklin Nóbrega
Infraestrutura e Banco de Dados
Planae Tecnologia da Informação
(+55) 14 3106-3514
http://www.planae.com.br

Patrick Vedrines wrote:
Hello Gustavo,

Your question seems to say that you suspect a disk issue, and a few hours later, Simon told me "Sounds like your disks/layout/something is pretty sick".
To be clear in my own mind about it, I've just copyed (time cp) the "aggregate" table files (4 Gb) from one disk to an another one: it takes 22 mn !(3 Mb/s).
That seems to demonstrate that Postgres is not the cause of this issue.
I've just untrusted to my system engineer the analysis of my disks...

In case I would have to change my disks, do you have any performance figures related to the types you mentionned (reiserfs vs ext3) ?
I don't use RAID since the security is not a concern.

Thank a lot for your help !

Patrick

Hi Patrick,

How is configured your disk array? Do you have a Perc 4?

Tip: Use reiserfs instead ext3, raid 0+1 and deadline I/O scheduler in kernel linux 2.6