Hi all,
Just hoping to open a general discussion and get a few ideas how network designers are dealing with the apparent conflict between multi layer network design and the desire for flat vlans to faciliate vMotion. And how people manage DR for layer 3 connected data centres.
Due to increased use of virtualization these days, many network designers are looking to extend their Layer 2 domains across multiple data centers to allow transperent vMotion. This seems to conflict with best practice multi layer network design that requires small layer 2 domains (to reduce stp risks).
I am working on a couple of network designs for DR sites. We are using Veeam to replicate servers from the primary datacentre to the DR site. For one project we have only a layer 3 (routed) link between data centres, and for the other we do have a layer 2 link between data centres.
Between the two data centres with the layer 2 link, we can vmotion a server from one DC to the other transparently. So our DR is fairly seemless.
For DR between the two Datacenters where we only have a layer 3 link, we have a nice multi layer design running OSPF. This means that in a DR senario out plan is to:
shutdown the Switch Virtual Interface (SVI) for the network that the VM was on at the failed site (if its still operating)
Turn on a SVI with the same IP and network at the DR site.
Turn on the replicated VM at the DR site
Cross fingers and OSFP will tell all our clients that are reliant on the failed VM where the replicated one was turned on.
The problem with the above is that its a manual process - I call it "subnet vMotion". Does anyone else have any ideas on how they handle this sort of DR setup neatly?
One idea I had was if Cisco and VMware (and the other virtulisation crowds) could develop something so that within Vsphere you could choose to "vMotion" an entire vlan (subnet). All single homed hosts with an interface in that vlan could move, and at the same time do talk to the routers/switches and disable the vlan on the source router and enable on the destination. There would be a bit of an outage while routing protocols converged, but it would save allowing the server guy having to do a lot of manual networking stuff. Or perhaps its already possible to do something like this?
Cheers, Simon.
