[Top] [All Lists]

RE: blocksize question

To: "'Eyal Lebedinsky'" <eyal@xxxxxxxxxxxxxx>
Subject: RE: blocksize question
From: "Mike Young" <myoung@xxxxxxxxxxxxxxxxxxx>
Date: Wed, 31 Dec 2003 21:28:48 -0800
Cc: "'linux-xfs list'" <linux-xfs@xxxxxxxxxxx>
In-reply-to: <3FF36DF1.C6973D7@xxxxxxxxxxxxxx>
Sender: linux-xfs-bounce@xxxxxxxxxxx

I have not tried the dd test yet.  I'll be getting to that tomorrow.  On a
dual 3.06 Xeon, I received a performance spread of 7MB/sec to 29MB/sec using
smbtorture across GbE using a crossover cable.  My client was a 1.4GHz P4.
The server also had 512MB.  The 29MB/sec was at 12 clients.  After that, the
output was a consistent 21.xxMB/sec up to 53 clients. I did this test using
an external log, which I have found to be much faster on such testing than
using the internal log.

However, I did the same test with an LSI controller and obtained over
40MB/sec at 53 clients.  I'm trying to also repeat the test using MD RAID.
In the past, I've had MD RAID doing over 50MB/sec on the same test.  But I
do want to repeat just to be sure.

Also, I have some messages over to the 3Ware folks to see if they can help
on this.  It sounds like the external journal should help you some.  

Any idea what the block size is on ext2 system?  I'm wondering if that may
be the issue.  The default you're using on your xfs system is probably 4K.


-----Original Message-----
From: linux-xfs-bounce@xxxxxxxxxxx [mailto:linux-xfs-bounce@xxxxxxxxxxx] On
Behalf Of Eyal Lebedinsky
Sent: Wednesday, December 31, 2003 4:47 PM
To: Mike Young
Cc: 'linux-xfs list'
Subject: Re: blocksize question

Mike Young wrote:
> Hi Eyal,
> I'm currently doing some performance testing with 3Ware and some other
> options on SuSE 9.0 and using XFS.  What kind of performance are you
> And what tests are you using?

We have a 7500-12, The raid5 is 8 disks + hot spare (XFS mounted
on /backup). We attempted to add a raid0 with two disks but this
seems to be too much load. This two disk raid0 was software raid
but we kept getting io errors
        {DriveReady SeekComplere Error}

The test is simply reading or writing (dd) a large file. For example:
        10GB /dev/zero   -> /backup     3m57 (43 MB/s)
        10GB /backup     -> /dev/null   7m20 (23 MB/s)
We do each 'dd' twice and take the second copy as the result. The
machine is otherwise idle.

This copying is a good measure for us since this machine is an online
backup service and it will mostly have large files copied to it and
from it.

Each disk alone on an IDE cable does about 55MB/s (hdparm, not real

Here are the results for the raid0 on the 3ware (/dev/sdb, ext2
on /ssaback):
        10GB /dev/zero   -> /ssaback    3m52 (44 MB/s)
        10GB /ssaback    -> /dev/null   2m42 (63 MB/s)

BTW, the XFS uses an internal log. I wonder if we should move to
an external log?

The machine is a dual AMD MP 2100+ with 1GB RAM.

Eyal Lebedinsky (eyal@xxxxxxxxxxxxxx) <http://samba.org/eyal/>

<Prev in Thread] Current Thread [Next in Thread>