Re: Antw: Re: [suse-sles-e] Re: catastrophic disk-performance on SLES10 with SATA drives

From: Markus Roeffen (markus.roeffen_at_iws.uni-stuttgart.de)
Date: Wed Apr 04 2007 - 12:23:56 CEST


Message-ID: <46137CBC.3080104@iws.uni-stuttgart.de>
Date: Wed, 04 Apr 2007 12:23:56 +0200
From: Markus Roeffen <markus.roeffen@iws.uni-stuttgart.de>
Subject: Re: Antw: Re: [suse-sles-e] Re: catastrophic disk-performance	on SLES10 with SATA drives

Hello Andreas,

I've made a few benchmarks with the tool "bonnie" (
http://www.textuality.com/bonnie/ ) and just wanted to post the results.
I've taken the RAID 5 SCSI results with ext3 at
http://cc.jlab.org/docs/scicomp/benchmark/bonnie.html as reference result.

Reference system (according to jlab.org):

    * Processors - Dual Intel Pentium III 1.26 GHz
    * Memory - 512 MBytes 133 MHz SDRAM ECC
    * Motherboard - Intel SDS2
    * RAID Controller - Mylex eXtremeRAID 2000 with 64 MBytes cache
    * Disk Drives - Seagate ST373405LC (Qty. 7)
    * RAID Configuration - RAID-5 (4 disks on channel-0 and 3 disks on
      channel-1)
    * File System = Ext3
    * OS - Red Hat 7.2, Linux Kernel 2.4.18-lkpc-4

So here's what I got:

-------Sequential Output-------- ---Sequential Input-- --Random--

             -Per Char- --Block--- -Rewrite-- -Per Char- --Block--- --Seeks--

Machine MB K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU K/sec %CPU /sec %CPU

EXT3 1024 14452 92.6 14325 11.2 12192 8.5 13508 81.9 73031 19.8 673.4 4.5

intern 1*1024 52313 94.4 67747 13.4 31852 7.2 50010 81.3 89308 8.2 714.8 0.9

extern 1*1024 24031 44.4 37296 9.9 27341 7.0 52420 85.5 94305 9.5 961.2 2.1

EXT3 is the old PIII-system on jlab.org.

intern is our internal RAID 1 SCSI
extern is our external RAID 5 SCSI

Best regards
Markus

matilda matilda wrote:
> Hi Markus,
>
> just one comment for today.
> I've never tested it, but I can imagine that the elevator mode could
> have an impact. SuSE comes with elevator=cfg as standard. In this mode
> the kernel tries to reorder write requests. This could be counterproductive
> if you have a hardware raid with buffer memory and own logic of summing up
> write/read requests.
>
> I found http://www.redhat.com/magazine/008jun05/features/schedulers/
> to be really interesting.
>
> Probably you have time and fun trying it.
>
> Best regards
> Andreas Mock
>

-- 
-------------------------------
Markus Roeffen
Institut für Wasserbau IWS
Universität Stuttgart
Pfaffenwaldring 61
70569 Stuttgart
Tel: +49(0)711/685-67010
Fax: +49(0)711/685-67020
Mail: markus.roeffen@iws.uni-stuttgart.de


This archive was generated by hypermail 2.1.7 : Wed Apr 04 2007 - 14:27:54 CEST