Avoiding Xen suspension at source after migrate.

Avoiding Xen suspension at source after migrate.

[NT_Mail List]

Xen Developers:

Today, we can move a VM from point A to point B. The
VM ends up at point B. After that, all resources are
reclaimed at point A. Is there a way we could that,
after the migration, the VM in A 
continues to run. We end up with two VMs, one in A and
one (its clone) in B?

I assume that today's Xen doesn't have any feature to
do this (e.g., xm clone). I wonder however if there is
a specific place in the Xen code that we can comment
out to make this trick work (intuitively, 
the clone operation requires less code than the
migration).

Do you have any other recommendation/present work that
can making this cloning possible.

thanks,
Chetan










__________________________________________________
Do You Yahoo!?
Tired of spam?  Yahoo! Mail has the best spam protection around 
http://mail.yahoo.com 

Re: Avoiding Xen suspension at source after migrate.

[Daniel P. Berrange]

On Thu, Sep 14, 2006 at 02:18:05PM -0700, NT_Mail List wrote:
> Xen Developers:
> 
> Today, we can move a VM from point A to point B. The
> VM ends up at point B. After that, all resources are
> reclaimed at point A. Is there a way we could that,
> after the migration, the VM in A 
> continues to run. We end up with two VMs, one in A and
> one (its clone) in B?
> 
> I assume that today's Xen doesn't have any feature to
> do this (e.g., xm clone). I wonder however if there is
> a specific place in the Xen code that we can comment
> out to make this trick work (intuitively, 
> the clone operation requires less code than the
> migration).

In Xen perhaps it would require less work, but the implications for the
guest OS are definitely non-trivial & the vast majority of the work I'd
say.

You need them to change their IP address(s) and MAC addresses - but
what do you do about processes with active network connections ?
You'll also need to allocate a new UUID to the guest VM (and restart
any apps which hook off the UUID within the guest). Not quite sure
how you'd signal to the guest OS to do this work, but would need to
make sure the 2 vms don't both transmit with same MAC/IP details
until one of them re-configured.

The hardest bit of all is probably the filesystems, in particular the
root filesystem. If you're using LVM you could have XenD create a
writable snapshot for one of the VMs - basically any way to stop the
two OS writing to the same root filesystem. One simpler approach
would be to run off a read-only filesystem eg the Fedora Stateless
Linux project.

Any this doesn't even consider the implications for specific applications.
What does it mean if you fork a VM with an Oracle database ? Perhaps these
are exactly the kind of problems you're interested in experimenting with 
though :-)

Regards,
Dan.
-- 
|=- Red Hat, Engineering, Emerging Technologies, Boston.  +1 978 392 2496 -=|
|=-           Perl modules: http://search.cpan.org/~danberr/              -=|
|=-               Projects: http://freshmeat.net/~danielpb/               -=|
|=-  GnuPG: 7D3B9505   F3C9 553F A1DA 4AC2 5648 23C1 B3DF F742 7D3B 9505  -=|