xfs
[Top] [All Lists]

Re: linux software raid5 on lvm: sunit/swidth required?

To: xfs@xxxxxxxxxxx
Subject: Re: linux software raid5 on lvm: sunit/swidth required?
From: Stan Hoeppner <stan@xxxxxxxxxxxxxxxxx>
Date: Tue, 18 Jun 2013 02:03:13 -0500
Delivered-to: xfs@xxxxxxxxxxx
In-reply-to: <57FC0B2322184044812B39B9440D8AF662898079AA@xxxxxxxxxxxx>
References: <57FC0B2322184044812B39B9440D8AF662898079AA@xxxxxxxxxxxx>
Reply-to: stan@xxxxxxxxxxxxxxxxx
User-agent: Mozilla/5.0 (Windows NT 5.1; rv:17.0) Gecko/20130509 Thunderbird/17.0.6
On 6/17/2013 10:06 AM, Thomas Molszich wrote:
> Hello all,
> 
> At present, does one still need to calculate and pass the sunit/swidth to 
> mkfs.xfs or does xfs calculate
> All this on its own?
> Xfsprogs 3.1.7, linked against libblkid, kernel 3.2.48 (Ubuntu 12.04 LTS), 
> 6x3000GB Hitachi 4k sectors
> with 512b emulation, partitions and lvm 4k aligned already.

Your subject says Linux software RAID5, i.e. the md driver.  But you
don't mention md in your msg body, only "partitions and LVM 4K aligned
already".

If you layer LVM atop a striped md array then mkfs.xfs cannot see the
geometry of the md array.  LVM is not aware of it because LVM isn't
providing the striping function, and it does not perform "pass through"
of the md geometry.  So in this case you must manually align XFS,
assuming your workload will benefit from alignment.  It may not, in
which case using the default 4K alignment would be better.

mkfs.xfs will auto configure stripe geometry if directly targeting an md
striped array.  It will not with LVM atop the striped array.

-- 
Stan

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