[Sheepdog] Config for dedicated storage network and NIC bonding?

MORITA Kazutaka morita.kazutaka at lab.ntt.co.jp
Fri Jul 29 19:36:12 CEST 2011


At Thu, 28 Jul 2011 14:01:38 +0100,
Matthew Law wrote:
> 
> On Wed, July 27, 2011 7:10 pm, MORITA Kazutaka wrote:
> > Yes.  Specify the ip address on the bonded interface to
> > totem.interface.bindnetaddr in corosync.conf.
> >
> > It should work, but I'm not sure whether it increases throughput.  It
> > depends on your environment.  If your disk is a bottleneck, you
> > wouldn't see any improvement.
> >
> >>
> >> I have seen mention of Sheepdog support in OpenStack.  Is this available
> >> now?
> >
> > Yes.  You can use Sheepdog from nova-volume in OpenStack.
> 
> Thanks. I shall try different bonding options on the hosts and switch and
> assess the performance.  I don't think the disks will be the bottleneck as
> each host has a 4 x disk RAID0 dedicated to sheepdog.
> 
> How far off being regarded as 'production ready' is sheepdog?  Is there
> anything you would like tested, test cases that require writing, running
> or anything I could do to help?

Thank you.  The test we can use for Sheepdog now is only qemu-iotests.
It is a client-side test and it tests only VM's I/Os against Sheepdog
volumes.  I think we need a test framework for a server-side code
before using production purposes.  I suspect there are some bugs
around Sheepdog data recovery and node membership management.  It
would be really nice if somebody create a framework to test those
things.

The other thing we need to do now is to create a recovery tool.  Some
users reported that their cluster became in an inconsistent state
because of Sheepdog errors or wrong shutdown.  We need something like
a fsck for Sheepdog.

Thanks,

Kazutaka



More information about the sheepdog mailing list