On 06/19/2012 08:22 PM, Michael Roth wrote:
On Tue, Jun 19, 2012 at 11:34:42PM +0900, Takuya Yoshikawa wrote:
On Tue, 19 Jun 2012 09:01:36 -0500
Anthony Liguori <anthony@xxxxxxxxxxxxx> wrote:
I'm not at all convinced that postcopy is a good idea. There needs a clear
expression of what the value proposition is that's backed by benchmarks. Those
benchmarks need to include latency measurements of downtime which so far, I've
not seen.
I don't want to take any postcopy patches until this discussion happens.
FWIW:
I rather see postcopy as a way of migrating guests forcibly and I know
a service in which such a way is needed: emergency migration. There is
also a product which does live migration when some hardware problems are
detected (as a semi-FT solution) -- in such cases, we cannot wait until
the guest becomes calm.
Ignoring max downtime values when we've determined that the target is no
longer converging would be another option. Essentially having a
use_strict_max_downtime that can be set on a per-migration basis, where
if not set we can "give up" on maintaining the max_downtime when it's
been determined that progress is no longer being made.
There is no need for a new parameter. Management software like
ovirt/virt-manager can track the mount of pages-to-migrate left and if
the number start rising, realize that the current max limit won't
converge and either increase the number or cancel the migration.
Although I am not certain whether QEMU can be used for such products,
it may be worth thinking about.
Thanks,
Takuya
--
To unsubscribe from this list: send the line "unsubscribe kvm" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
--
To unsubscribe from this list: send the line "unsubscribe kvm" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
