[Linux-cluster] Software RAID support by GFS

[Lon Hohberger]

On Wed, 2006-12-06 at 09:57 -0500, Josef Whiter wrote:
> Hello,
> 
> Software RAID has the same limitation as LVM1 as it's not cluster aware.  Even
> with LVM2 you have to use CLVM to make sure the appropriate cluster wide locking
> takes place before touching the metadata on the disk.  Software RAID has no
> cluster awareness, so you could run into the same problem you would have if you
> mounted an ext3 filesystem RW between two nodes, you would end up with
> corruption.  This isn't a limitation with GFS, but more a limitation with md
> RAID, as its not cluster aware.  Thank you,
> 

I *think* this is because write ordering between parts of a md-raid set
isn't guaranteed...  I've actually forgotten how the corruption occurs.
(d'oh!)

e.g. Two guys write to the same md-block at the same time.  One writes
in {1 2 3 4 5} order, the other writes in reverse order - the stripe for
that block is now corrupt.

... not to mention cache-coherency problems with md-raid.

-- Lon
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 189 bytes
Desc: This is a digitally signed message part
URL: <http://listman.redhat.com/archives/linux-cluster/attachments/20061206/a8e7dcb0/attachment.sig>