Open Stack

Lingxian Kong wrote:

>Actually, in the scenario of NFV, all the rules or behaviors of the physical world will apply to that in the virtual world, right?

>IMHO, despite of the scenarios, we should at least guarantee the consistency of creating vms with nics and attaching nics .

I'll need to think about that a bit before I decide whether I agree. My gut response is to disagree with it as a blanket statement while allowing that it may apply in specific scenarios.
The point about PCI passthrough in another post was a good one. If the VM is managing physical NICs then LACP at the VM level would make sense. But if the VM is using virtualized NICs there's at least some possibility that the underlying connectivity from the vNIC to the physical network is going over an LACP bundle of multiple NICs handled at the hypervisor level. At least that's the way we're doing it.
Running LACP over multiple vNICs just seems wrong to me. Increasing availability doesn't simply mean adding two (or more) of everything. Sometimes adding more "things" reduces availability. There needs to be at least one specific failure scenario where thing 2 can be expected to continue working when thing 1 fails. If all the failure modes are correlated (i.e. whatever caused thing 1 to fail almost certainly would also cause thing 2 to fail simultaneously) then having one thing would be better than two.

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstack.org/pipermail/openstack-dev/attachments/20140127/ab9476b5/attachment.html>