[DRBD-user] nested lvm & dual primary

Colin cbex at xplornet.com
Sat Sep 6 23:17:39 CEST 2008

Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.

//forwarding this email to the mailing list//

OK Tom, I'm slowly getting the picture.  Maybe you (or anyone) could give me a 
boost, by telling me (us) how you would set up the disks/filesystems?  

2 node system
1 disk per node
Gb NIC w/crossover cable
100Mb NIC for the network

data encrypted on disk
a (large) portion of disk mirrored between nodes

for the purpose of:
hosting several virtual machines
where one node can host all machines during failover


On Saturday 06 September 2008 12:20:03 you wrote:
> On Sat, 6 Sep 2008, Colin wrote:
> > Tom, I think you are correct there.
> >
> > I guess a cluster-aware filesystem would only be needed if all the
> > virtual machines were on the same logical volume, and both nodes would be
> > running different VM's from that LV?
> as far as I know, you need the cluster aware filesystem if multiple nodes
> have the same filesystem mounted. So, for example, if you were using image
> files (yuck) stored on a single filesystem and handed out by each dom0 to
> guests on both nodes you would need it... as both dom0's have the
> filesystem mounted.
> That isn't to say that the cluster aware lvm version isn't a good thing...
> if both dom0's see the drbd resource, they better agree on how you've
> sliced it up into logical volumes... but the domUs using the LV slices can
> simply run a normal filesystem and get much better performance as they
> don't have to do any checking/locking to see if another node is messing
> with the same files.
> -Tom
> > I wonder if that kind of setup would be better or worse?
> > (Maybe more simple because only one LV is needed, but worse because of
> > the unnecessary  complication of using a 'distributed lock manager'?)
> >
> > Colin.
> >
> > On Saturday 06 September 2008 11:35:42 you wrote:
> >> On Fri, 5 Sep 2008, Colin wrote:
> >>> Well it came in to play when I had one large 'resource' configured
> >>> (Primary/Primary, if I didn't mention that yet), and I wanted to create
> >>> several 'logical volumes' on that resource, so that my multiple virtual
> >>> machines would have a defined sized box to live in.  This brought me to
> >>> the DRBD User's Guide, Chap. 10, Using GFS with DRBD.
> >>> On the page: Configuring LVM to recognize the DRBD resource, it states
> >>> that GFS uses CLVM.
> >>
> >> AFAICS, you do NOT need a cluster-aware filesystem. You're not going to
> >> have multiple nodes writing to the same filesystem at the same time. If
> >> you're given each virtual machine a chunk of space to call it's own,
> >> then that virtual machine will either be on one node or the other... and
> >> it will be the only writer to that chunk of space... yes/no?
> >>
> >> -Tom

More information about the drbd-user mailing list