> > You could also consider network-based block devices, e.g. NBD or iSCSI
> > to provide guest storage.
>
> (big snip)
>
> i'd admittedly neither considered this approach, nor even heard of
> ENBD! so, something new/interesting ...
>
> reading here, http://www.it.uc3m.es/ptb/nbd/, iiuc, this lends itself
> to implementation on a separate box, no?
NBD is, IIRC, not safe to "loop back". So it's not safe to connect to and
mount a NBD device which is being served out by the local machine. This is
to avoid deadlocks. I assume this is what you're referring to?
Really the restriction should be phrased as "it's not safe to connect to and
mount a NBD device which is being served out by the local *OPERATING
SYSTEM*". So as long as the NBD server and client are in separate domains,
it's safe to put them on the same physical host.
> or, are you suggesting deploying an NBD server in/as a DomU?
Well, the NBD script shipped by Xen allows you to have an NBD server wherever
you want. It could be a domU on the same machine, or it could be a separate
host. The point of the script is that it sets up an NBD device *in dom0*
that connects to the NBD server. It then re-exports that as a standard Xen
VBD to the guest.
The guest sees a normal disk and has no idea that it's actually network
transparent. You can shut the guest down, move the disk to an NBD server on
another machine, then boot the guest again without making any changes to the
guest itself.
In actual fact, I suspect it is actually safe to export a block device via NBD
in dom0, then map that NBD device in dom0 and export it to a domU (at least
for a PV domU. For HVM I might need a bit more caffeine in the system to
decide whether it's a good idea but I suspect might be OK). This should be
safe because dom0 wouldn't actually mounting the device, just passing it
through. Doing this would be rather pointless, unless you wanted to
occasionally migrate the guests off the original machine but leave the
storage there. I don't imagine it'd be good for performance either, I'm just
making you aware of what could be possible (even if not recommended!).
> thanks for the idea(s)! your well written comment does a nice job of
> intro as to capability; understanding differences wrt
> NFS/OpenFiler/other ntwk storage requires some add'l reading on my
> part ...
At this point it's probably worth asking what exactly you're hoping to achieve
so I can direct my advice a bit better. If you want some kind of shared
filesystem between domains that multiple domains can access concurrently, you
need either a network filesystem (NFS or CIFS) or a cluster filesystem (GFS
or OCFS2). If you want a way of centralising block storage on a server
there, you want iSCSI or *NBD. And so on :-) There are lots of options here
but if you tell us a bit more we might be able to suggest some further
options.
Cheers,
Mark
--
Push Me Pull You - Distributed SCM tool (http://www.cl.cam.ac.uk/~maw48/pmpu/)
_______________________________________________
Xen-users mailing list
Xen-users@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-users
|
Home