Re: [Xen-users] Re: XEN Images Storage

["Daniel Asplund" <danielsaori@xxxxxxxxx>]

Have a look at using DRBD, Heartbeat and XEN with two physical machines.
That is the setup I'm currently testing and so far it works great. At
the moment I use old hardware(from the P3 era) with old discs and only
100Mbps NICs, but still a live migration of a DomU is hardly
noticeable.

I had a go first with the Redhat cluster suite, but it doesn't run
well with XEN and DRBD. I had to go back in versions to get it running
but then I lost important functionality.

With version 0.8 of DRBD there is support for XEN live migration. I
really like the way XEN takes care of the my DRBD resources, I have no
script for this, it is all up to XEN.

I then use Heartbeat to monitor the status of each host that will
automatically migrate DomUs to the other host if needed. Another
recommended thing is to use a STONITH device so Heartbeat can reset of
the other host in case of a failure, have not tried that yet.

Cheers, Daniel


>
> > I want to build on Open Source tools a high availabitiy server to store
> > my images, someone with suggestions? experiences plz. really aprreciate.
> >
> > Started to try Gluster but yesterday discussion pointed me very
> > important things.
>
> For Image storage you don't need a Cluster Filesystem. It even maybe a
> performance bottleneck, as any Clusterfilesystem will need write locking over
> the net.
>
> What you really need is some kind of shared storage. That could be
> Fibrechannel, Iscsi or (today not any more common) SCSI.
>
> The shared storage must have redundant paths to be really high availability.
> So that means: Redundant paths to the storage as well as redundant raid
> controllers. (For the FC example:  Dual HBA's, Dual FC Switch, Storage with
> Dual Controller). For the path discovery use dm-multipath.
>
> What I'm doing in the front:
>
> Three server (all with dual FC HBA) with Scientific Linux 5.1 (a RHEL clone
> like CentOS). The servers are members of a cluster (of course you need
> servers with a BMC and IPMI over Lan for the fencing).  On the clusternodes I
> run clvmd (Cluster Logical Volume Manager) and rgmanager (Resource Manager).
>
> The images are just Logical Volumes, they will not be mounted on the hosts.
> The cluster locking is only needed for the lvm commands, so its no
> bottleneck. Any node sees the same images, so on failover another node could
> immediately restart the failed domU.
>
> rgmanager handles nicely the failover of domU's.
>
> If it's enough availability when you're domU's are up again after a host
> failure after a few 10's of seconds, then that's it.

_______________________________________________
Xen-users mailing list
Xen-users@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-users