Lustre discuss - Jul 2007 - MDT Failover

Hello,

I wondered if it''s possible to have two MDTs in failover who write
their data to two different devices? One MDT1 active as normal, the
other MDT2 would be passive but still writing it''s metadata to
it''s
own device. So that in a case of failure the first MDT would go down
normally, the second one would come up with it''s own partition, which
should be nothing more than a exact copy of the metadata device of
MDT1?
It would be great, because we have not one large Server room in our
building, but a lot of small ones, each with an air conditioned server
rack, and we still want to use our clusterfs, even if the "whole rack"
with the  MDT1 fails.

I looked trough the manual and found the "failover HowTo" and I only
figured out, how to use two MDTs with the same device, which wouldn''t
be  a real "failover" in our setup, or do miss something?

If there is no possibility, is there some way of "workaround" like
using iscsi or something like that to update the second MDT-device so
that if the first one fails, the second is up to date with it''s own
device?

Sorry to bother you with this questions but I found nothing documented
for this kind of setup.
And of course, sorry for my bad english

Greetings
Tobi

On Wed, 2007-07-18 at 00:03 -0700, Tobi wrote:
> Hello,
Hi,
> I wondered if it''s possible to have two MDTs in failover who write
> their data to two different devices? One MDT1 active as normal, the
> other MDT2 would be passive but still writing it''s metadata to
it''s
> own device. So that in a case of failure the first MDT would go down
> normally, the second one would come up with it''s own partition,
which
> should be nothing more than a exact copy of the metadata device of
> MDT1?
You want DRBD http://www.drbd.org/
> I looked trough the manual and found the "failover HowTo" and I
only
> figured out, how to use two MDTs with the same device, which
wouldn''t
> be  a real "failover" in our setup, or do miss something?
Well, in terms of redundancy of the actual storage device, true.  Even
if the disk array were using a RAID such as RAID 1 5 or 6, it would not
be immune to a piano falling on it.  With DRBD you can achieve that
level of redundancy assuming you have enough bandwidth between the two
physically separate disks to keep up with the data throughput of the
disk.
> If there is no possibility, is there some way of "workaround"
like
> using iscsi or something like that to update the second MDT-device so
> that if the first one fails, the second is up to date with it''s
own
> device?
DRBD.
> Sorry to bother you with this questions but I found nothing documented
> for this kind of setup.
> And of course, sorry for my bad english
No worries.  Your English is a far cry better than my German.

b.