On Mon, 2007-01-22 at 11:34 -0200, Christopher G. Stach II wrote:
> Reinhard Brandstaedter wrote:
> > On Fri, 2007-01-19 at 11:18 -0200, Christopher G. Stach II wrote:
> >> You have a few basic choices:
> >>
> >> 1. Export a single LV from dom0 as a full disk, allowing your domU to
> >> see the LV as an entire disk and use partition tables.
> >> 2. Export each single LV from dom0 as partitions, allowing your domU
> > to
> >> only see those partitions.
> >> 3. Export the VG device from dom0, allowing your domUs to see all LVs.
> >>
> >> Since you're using a SAN, you may be aiming to support multiple Xen
> >> boxes on the same device(s). If so, you're probably going to want to
> >> run CLVM. The question is, "Where?" If you use #1 or #2, your dom0s
> >> will have to participate in the cluster. If you use #3, you push the
> >> clustering to the domUs (where I think it belongs.) #3 is less
> > secure,
> >> however.
> >
> > That's exactly what I want to do - booting several VMs from one
> > read-only rootfs (one LV). But I think I will run CLVM on the Dom0s and
> > use method #2. I haven't used CLVM yet, but I think it's easier to
> > configure it on 2-3 Dom0s than on every DomU?
>
> It's "easier" (if you've done it more than once, it's not that
> difficult, but the first time...) if you run it on fewer dom0s, but it
> also puts all of the VMs at risk of going down if your dom0 gets fenced
> and reset. It also puts greater load on the dom0s, which could impact
> VM performance quite a bit (like if you're using default SEDF params.)
>
> You _could_ also use LVM in the dom0s without the cluster capabilities,
> but you won't see any LV changes across the cluster until you reset the
> machine(s). As with all of the other cluster suite stuff, there's also
> the risk of it exploding in your face. :)
The scenario Chris described happened to me (unfortunately) a few times
until I really began simplifying things. The NFS / Open SSI classic way
of doing things does apply here, and a many-read-no-write volume
shouldn't be too risky.
Don't use a cluster FS for something you could pull off with NFS, no
matter how much easy it may seem to make your 'down the road' ideas work
better. If you need a cluster FS, I highly recommend going with ocfs2 vs
the RH cluster suite.
It almost always every time blows up in your face somehow, and I'm still
hearing nasty things about a GFS inode DOS vulnerability. I'm not saying
that still exists, I'm just pointing out some would rather risk it than
try upgrading GFS. :)
Best
--Tim
>
_______________________________________________
Xen-users mailing list
Xen-users@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-users
|