At Mon, 23 May 2011 12:24:22 +0200,
Jérémie Tarot wrote:
> Hi,
> 
> This is a very interesting scenario...
> 
> 2011/5/23 MORITA Kazutaka <[email protected]>
> 
> > At Sun, 22 May 2011 11:25:44 +0200,
> > Gandalf Corvotempesta wrote:
> > > 2011/5/21 MORITA Kazutaka <[email protected]>
> > > >
> > > > The patched version is also available at:
> > > >  git://sheepdog.git.sourceforge.net/gitroot/sheepdog/qemu iscsi
> > > >
> > > > See also
> > > >
> > https://sourceforge.net/apps/trac/sheepdog/wiki/General%20Protocol%20Support
> > >
> > >
> > > If I understand correctly with this patch i'll be able to export a whole
> > > sheepdog cluster in a single iSCSI block device and then mount it on a
> > > different server that will run the VMs ?
> > > Like a distributed SAN, for example:
> >
> > You can export a Sheepdog volume (not a whole sheepdog cluster) as an
> > ISCSI LUN.  Please note that the iSCSI support is experimental.  I/O
> > is not asynchronous (poor performance), and the export configuration
> > is not shared among Sheepdog nodes.
> >
> > >
> > > 3 sheepdog node merged in a single cluster and then exported to a single
> > > iSCSI device.
> > > In front of them, will be 5 debian boxes that mounts the iSCSI device and
> > > run VM from it ?
> > >
> > > If one debian box will die, other 4 will take care of the lost VM by
> > running
> > > them from the iSCSI devices archieving fault tollerance.
> >
> > Automatic failover of VMs is not supported.  I think it is out of
> > scope of the storage system.
> >
> > >
> > > If one sheepdog node will die, other sheepdog will takeover it and move
> > the
> > > iSCSI target to an available node?
> >
> > No.  You need to start the iSCSI target on the available node, and
> > export Sheepdog volumes with the same configuration manually.
> >
> 
> In this given setup:
>    * Can we export the same sheepdog volume from many/all sheepdog cluster
> members at the same time ?

Yes.  If you run iscsi targets on multiple machines, it should work.

>    * Can't we use multipathing to connect to the targets to gain fault
> tolerance ?

Probably, active-standby mode would work, but I haven't tested at all.

The restriction of Sheepdog is that one Sheepdog volume cannot be
accessed from multiple VMs at the same time.

Thanks,

Kazutaka
-- 
sheepdog mailing list
[email protected]
http://lists.wpkg.org/mailman/listinfo/sheepdog

Reply via email to