> > > Has anybody looked at clustering support for Software
> RAID (ie. in a
> > > shared disk environment: shared scsi or FC). I guess
> this may have
> > > come up on the list before but I didn't find anything
> in the maillist
> > > archives (LVM guys once discussed this but not sure if
> there is any
> > > existing support for this)
> > >
> > Have a look at http://www.complang.tuwien.ac.at/reisner/drbd/

> nope drbd handles data replication, what Michael is looking for is
> a distributed lock system. LVM guys are thinking of
> something for future
> releases, but at the moment i don't know of anything existing.
> maybe some creative use of heartbeat?

Yeah, exactly. I can use Software RAID at the moment with my shared
SCSI setup although I need to down one of the nodes to make topology
changes - and I can't use hot spare (to avoid both hosts rebuilding
the same array) - so I also need to down one of the nodes if I want to
do a rebuild.

What i'd ideally like is one big RAID5 md over 7 disks (in my shared
enclosure) and use lvm to carve this up into a number of smaller
logical volumes - with each LV only being used by one of the nodes in
the cluster at a time - this can work now but doesn't handle topology
changes or rebuilds while both nodes are up.

I am using kimberlite http://oss.mclinux.com/kimberlite/ - this
manages the exclusive access to individual partitions/LVs making sure
they are only mounted one one of the cluster nodes.

GFS unfortunately is not suitable as my whole cluster will go down if
the node running the lock server dies (I understand they are working
on eliminating this single point of failure). And it also suffers from
the same problem stated above ie. if you want to run GFS ontop of
Software RAID, you need the mds active on both nodes.

At the moment, I might just stick with hardware RAID - but I think
this would be quite neat as it would cut the cost of a cluster by a
bit.

~mc

-
To unsubscribe from this list: send the line "unsubscribe linux-raid" in
the body of a message to [EMAIL PROTECTED]

Reply via email to