OTV is Dead. Long Live EVPN-VXLAN!
- Written by: Wes Sudman
Several companies I have encountered in the past have dual Data Centers but are not really taking advantage of this design. Whether the goal is active/active or active/standby these two Data Centers have to communicate with each other and this is where the vast majority of networks differ.
Data Center Interconnect (DCI) links exist as our high-speed form of communication between these Data Centers. When it comes to the DCI there are several technologies that can come into play. Traditionally our connectivity was Layer 3 between these sites which was fine to route between them but what about Layer 2? With the emergence of vMotion, companies looked to take advantage of these links to migrate workloads between Data Centers. One caveat was that this required Layer 2 connectivity. This is where OTV came into play.
But can’t we just utilize a Layer 2 DCI link? This presents another problem: Spanning tree. OTV was designed by Cisco not only to solve this problem of extending Layer 2 over a Layer 3 network but to also limit the Spanning tree topology to a single DC. Another problem it solved was for FHRP to be segmented as well so we could have network gateways in both DCs. Sounds great right? Initially it checked a lot of boxes to satisfy the needs at the time. So why is it dead? Several reasons. First is that its proprietary. You must be running Cisco hardware and it is only available on ASRs, and Nexus 7Ks. Second is that you can only “load balance” across both links by VLAN IDs. Your “even” VLANs ride one link while your “odd” VLANs ride another. This did help split up the traffic but was hardly an efficient way of doing so. And now with 80 percent of DC traffic being east/west, better efficiency is needed. The final reason is convergence. OTV doesn’t handle disruptions well and utilizes a 240 second timer before traffic is put back on the disrupted link. Four minutes seems an eternity when it comes to network down time! So, what’s the answer?
VXLAN EVPN. It’s an open standard. Can be run across multiple vendors. Provides more efficient routing/switching and converges quickly. It’s everything OTV is and isn’t. Well, then what’s the downside? Complexity. And this is where vendors have stepped in with their own solutions to help simplify the process. Cisco’s ACI or DCNM, Juniper’s Apstra, BigSwitch, and countless others have all leveraged VXLAN EVPN for their “Fabric” solutions. While I can’t with a straight face say that Cisco’s ACI is simple it does allow all configuration to be completed in a GUI which relieves the engineer from having to go into the weeds of VXLAN EVPN Fabric. While it is definitely beneficial to understand the logic of how this technology works for troubleshooting it isn’t entirely necessary.
So, is OTV completely dead? Not yet. But it does have one foot in the grave. As more and more vendors release their VXLAN EVPN solution it has essentially turned OTV vs the World. So, it may be time to take a look at your network to make sure you aren’t left out in the cold with a dying technology. If you are interested in learning more get in contact with your local HBS account manager and let’s have a talk.