> OS: SLES 11 SP1 with onlin updates + HAE SP1
> if I cluster a FV(win 2k8) XEN machine, in case of migration(or the
> node running the vm goes down).. may we expect a downtime(restart of vm
> on the other node) ? or does it resume on the
> surviving node without a downtime(or with negligible downtime)?
> sorry if I am asking this question in wrong place
You've posted in the right spot
The answer to you're question depends on what you are doing/what is
happening... You mention both HA and Live migration in the situation
you are describing.
HA will kick in if your host fails.... but the way it kicks in is by
simply starting (cold boot as to say) the VM's that where running on the
Xen host that just failed.
Unless you can predict the Xen host is going to fail, the active state
of the VM's will fail just as hard as the Xen host. So it's high
availability as there is an automated process to (re)start the VM's as
quickly as possible on another Xen host that is still alive.
Live Migration on the other hand, is done between active Xen hosts...
and when setup correctly you should not notice any downtime - just a
little hickup while the VM hops over from one Xen node to the other.
This works by first having the Xen host that is running the VM you want
to migrate, to copy the active state (RAM, etc) of that VM to the other
Xen host... when that active state has been synced between the Xen hosts
(99%), the VM is put in a frozen state on the current Xen host... last
bits are synced over and the VM is unfrozen on the other Xen node.
Typically you'll only miss one en two seconds of connectivity and the
node will be back online without it (or the connected clients) being
aware of anything have happened.
The only solution I've worked with that can run a fully fault tolerant
VM (from a host perspective) is VMware vSphere. It is resource hungry
as you are actually running two sessions of the VM, each session on one
of the hosts where one session is active and the other is read-only (and
constantly syncing with the active one).
With this solution you can pull the plug on one host, and your VM will
stay running in an active state. Nice, but again quite resource hungry.
There could be other solutions that do something similar... I have not
worked with them.