xfs
[Top] [All Lists]

Re: XFS on RAID 10

To: Thomas Duffy <Thomas.Duffy.99@xxxxxxxxxxxxxxxx>
Subject: Re: XFS on RAID 10
From: Simon Matter <simon.matter@xxxxxxxxxxxxxxxx>
Date: Fri, 12 Oct 2001 08:00:44 +0200
>received: from mobile.sauter-bc.com (unknown [10.1.6.21]) by basel1.sauter-bc.com (Postfix) with ESMTP id 4F70157306; Fri, 12 Oct 2001 08:00:45 +0200 (CEST)
Cc: linux-xfs@xxxxxxxxxxx
Organization: Sauter AG, Basel
References: <1002857059.5390.11.camel@tduffy-lnx.afara.com>
Sender: owner-linux-xfs@xxxxxxxxxxx
Thomas Duffy schrieb:
> 
> I have setup a little (280G) xfs on RAID 10 (RAID 1 on RAID 0).  This

Some time ago I was playing with similar setup. IIRC I saw the same slow
resync speed. I then turned my RAID upside down and everything was okay.
I dont know how but since you said 'RAID 10 (RAID 1 on RAID 0)' I guess
you could just make it (RAID 0 on RAID 1) and it should be okay. In may
case sync speed was then ~10000-40000 w/o tuning. I'm interested to hear
what came out after.

> has 8 SCSI drives (4 per card).
> 
> Everything seems to work fine, expect the RAID resync craaaawwwllllsss.
> It goes at the minimum of 100k/s which should take a few months to
> finish syncing :)  It sometimes bumps up to 250 k/s, but never higher
> even with no system activity.
> 
> First off, does anyone know of a way to bump the sync rate up?  Could

I put this in my /etc/sysctl.conf
# Force RAID reconstruction to run with high priority
dev.raid.speed_limit_min = 5000

Usually you should not need this. If it does not sync fast while there
is no load on the box, something else is not good. But if you have
constant load on the box you really need the higher value than the
default (100).

> this be some sorta weirdness between XFS and MD that is causing the slow
> resync?  On a simple RAID 1 setup (2 disks) w/ XFS on the device, the
> resync is as speedy as the SCSI card allows when there is no other I/O
> which is what I would expect from MD.
> 
> Other info, I am using 64k chunk sizes on both the RAID 1 and RAID 0
> with 4k block sizes on XFS (obiviously because it on i386).  These are
> 10000 rpm ultra SCSI 3 drives, so they should scream.  I was able to dd
> a blank 200G file relatively fast using bs=64k on this device, so it
> seems to be ok with creating/copying files.
> 
> Thanks for any help,
> 
> -tduffy

-- 
Simon Matter              Tel:  +41 61 695 57 35
Fr.Sauter AG / CIT        Fax:  +41 61 695 53 30
Im Surinam 55
CH-4016 Basel             [mailto:simon.matter@xxxxxxxxxxxxxxxx]



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