The layer-2 appeal
Extending a VLAN between data centers lets a server move from one site to another without changing its IP address. Some legacy clustering and VM mobility tools expect this. It feels simple.
The hidden cost
A stretched layer-2 domain is a shared failure domain. A broadcast storm, a loop or an unknown-unicast flood at one site can spill into the other. Troubleshooting becomes harder because the problem has no clear boundary.
The layer-3 approach
Routing between sites gives each data center its own subnets and its own failure domain. Problems stay local. Routing protocols handle failover cleanly. Applications that are designed for distributed operation — using DNS, load balancers or service discovery — work well over layer 3.
The middle path: overlays
EVPN with VXLAN can provide layer-2 services for the few applications that need them, while the underlying network between sites is routed. You get controlled L2 extension without a single large broadcast domain.
How to decide
- Do your applications truly need the same subnet across sites? If not, route.
- If yes, can you limit the stretched VLANs to specific workloads?
- Do you have the operational skills to run an overlay?
Our recommendation
Default to layer 3, and use L2 extension only where an application requires it. Peer1IX DC-to-DC services support both, so you can choose per circuit.
