On Wed, Oct 24, 2012 at 2:33 AM, Richard Bin liu
<[email protected]> wrote:
> To realize seamless migration in wide range is one of our goals, but
> currently VM migration tools is likely to require: the distance between two
> places cannot be more than 50KM (or a larger value, but not too much). We
> know that disaster recovery point should be as far as possible from the
> local.

Hello, Richard, I am new to these technologies and discussion, but we
are working out how to implement VXLAN and including the concern you
mention above.  I think it is not specifically relevant to VXLAN and
friends.

Our plan deals with Internet->VM traffic through VTEPs located at the
multiple datacenters, all of which may be aware of Internet->VXLAN
mappings by way of the provisioning system plus ARP/MAC
learning/snooping.  VM->Internet traffic flows natively (after passing
through a filter in the hypervisor's VTEP) and is not affected by
triangular routing, because we will manipulate ARP and NDP traffic to
trick the guest OS into doing what we want -- sending its traffic to a
VRRP gateway that is present "everywhere."  It will not tunnel back to
a hub VTEP or require state information in the native layer-3 network
used to carry the Internet-bound traffic.

VM<>VM traffic within a customer's private L2 or L3 domain will
certainly have high latency if the two VMs are located far apart, but
this is a limitation of physics, not VXLAN.  I would welcome an
internet-draft accelerating the speed of photons through glass if you
can demonstrate a way to implement it.

-- 
Jeff S Wheeler <[email protected]>
Sr Network Operator  /  Innovative Network Concepts
_______________________________________________
nvo3 mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/nvo3

Reply via email to