From: Alexei_Roudnev (Alexei_Roudnev_at_exigengroup.com)
Date: Tue Aug 28 2007 - 00:09:31 CEST
Message-ID: <078501c7e8f6$f1bc0b20$7931a8c0@exigengroup.com> From: "Alexei_Roudnev" <Alexei_Roudnev@exigengroup.com> Date: Mon, 27 Aug 2007 15:09:31 -0700 Subject: Re: [suse-sles-e] Slow direct I/O performance SLES 10 SP1
I use directIO with SLES9 SP3 (and oracle) all the time, without any
problems. Why do you think that it's broken?
Most likely reason of the problems is HBA card , driver and switch (these
proproetary hardware and software never was stable - one step aside of the
trail and you have trouble). I'd suspect direcr and HBA card, and then
switch.
----- Original Message -----
From: "Bernd Melchers" <melchers@CIS.FU-Berlin.DE>
To: <suse-sles-e@suse.com>
Cc: "Andreas Mock" <Andreas.Mock@web.de>
Sent: Monday, August 27, 2007 1:23 AM
Subject: Re: [suse-sles-e] Slow direct I/O performance SLES 10 SP1
Same here with SLES9SP3, but better values with larger block sizes:
% uname -a
Linux 2.6.5-7.286-sn2 #1 SMP Thu May 31 10:12:58 UTC 2007 ia64 ia64 ia64
GNU/Linux
% sginfo -a /dev/sdd | fgrep 'Write Cac'
Write Cache Enabled 1
% sginfo /dev/sdd
INQUIRY reponse (cmd: 0x12)
---------------------------
Device Type 0
Vendor: FUJITSU
Product: MAW3300NC
Revision level: 0104
% dd_neu oflag=direct if=/dev/zero of=data.out bs=x count=y
y= some large enough values
x=4k: 532 kB/s
x=64k: 7.9 MB/s
x=256k: 26.6 MB/s
x=2M: 68.1 MB/s
x=6M: 80.0 MB/s
x=18M: 85.2 MB/s
disk is SCSI on qla12160, scsi write cache enabled
(same with write cache disabled)
direct i/o is broken with SLES9/SCSI
What about the block size dependency on SLES10?
mit freundlichen Grüßen
Bernd Melchers
---------------------------------------------------------------------
To unsubscribe, e-mail: suse-sles-e-unsubscribe@suse.com
For additional commands, e-mail: suse-sles-e-help@suse.com
---------------------------------------------------------------------
To unsubscribe, e-mail: suse-sles-e-unsubscribe@suse.com
For additional commands, e-mail: suse-sles-e-help@suse.com
This archive was generated by hypermail 2.1.7 : Tue Aug 28 2007 - 02:13:19 CEST