Re: AW: [Xen-users] Deploying redhat clusters under Xen 3.0.2

["Christopher G. Stach II" <cgs@xxxxxxxxx>]

Javier Guerra wrote:
> On Monday 21 August 2006 7:53 am, Christopher G. Stach II wrote:
> > would do differently.  First, don't let dom0 manage any of the storage.
> >   Use one of GNBD, iSCSI, AoE, etc. and dedicated private network cards
> > for each domain.  I wouldn't even bother bridging vifs.  Second, use the
> I see why not using vif-bridging for storage, but why not to manage storage in 
> dom0? having there the GNBD, iSCSI, AoE, device(s) and CLVM should work, and 
> it seems would be the most manageable setup.
> would you share your (bad) experiences with Dom0-managed storage?
Oh, it works, but not very well under high I/O load.  You want to 
dedicate that single physical processor to dom0 for that, but even then, 
you will end up with pcpu contention in the domUs (if you're running 
more vcpus than pcpus, minus the one for dom0.)  Eventually, they lag so 
much that CMAN drops them and they get fenced.  I'm sure that the credit 
scheduler will help, but I'm doubtful that it will help that much.  It's 
pretty clear to me that in such cases, you really should just let dom0 
manage the domUs and let the physical hardware do the work (network I/O 
to some block device and the other cluster nodes.)  I had been running 
this configuration in 3.0.2 since it came out.  It wasn't very happy.
Why would you want a choke between your block devices, anyway?

--
Christopher G. Stach II

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