Let's suppose that we have a common Clos EVPN/VXLAN-based DC fabric with IP gateways on leaves (anycast gateways). Also there is a lot of different IP VRFs inside a fabric where end hosts reside. What is an optimal way to connect all those VRFs to the MPLS-based backbone network (there is no EVPN at the backbone level, only IP VPNs).

I can imagine the following options:

  1. Option-A style. For each VRF create BGPv4 session between Border Leaf and DC Edge router. A lot of config overhead, poor scaling.

  2. Introduce DC Edge router into EVPN fabric, exchange EVPN routes with the fabric via BGP EVPN, create all necessary fabric VRFs at DC Edge router. In this option DC Edge router must perform MPLS-to-VXLAN transition and vice-versa. I'm not sure this is possible at the hardware level (DC Edge router is Juniper MX) and I cannot find any guides discovering this topic.


I've added the scheme to illustrate my goal.

I just want to connect VRFs A and B from the left to the same VRFs on the right with the following constraints:

  1. No need for stretching EVPN routes inside all Backbone network.
  2. No need for stretching IP VPN (MPLS) routes inside DC network.

I've pointed two options above to accomplish my goal. I prefer option number 2 (introduce DC Edge router into EVPN fabric and so on), but I'm not sure will it work or no. DC fabric and MPLS Backbone

  • Maybe you should clarify your goals. The options you propose make zero sense to me. When I read your post, the first thing I thought is, is this person unaware of route-reflection? and what remote things and boxes do they want to add to their EVPN domains?. This suggests you need to supply more information about your architectural goals. Nov 2, 2020 at 13:57
  • 1
    To add to Jeff’s point - the overhead of peering per-VRF is dependent on scale and change velocity. Peering a couple of dozen VRF’s once and then touching the config twice a year isn’t a huge deal. Thousands of VRF’s that see 15% churn per week drives a very different solution.
    – rnxrx
    Nov 2, 2020 at 16:40
  • @JeffWheeler, I've added more description in my question.
    – Anton M.
    Nov 2, 2020 at 19:52
  • You can operate EVPN domains on the right-most PE router in your diagram regardless of whether the other backbone nodes are aware of the EVPN topology. The DC Edge Router doesn't have to be involved unless you want it to be. The MPLS Backbone P node also doesn't. This flexibility is core to the concept of MPLS and related signaling protocols; it's meant to be usable this way. This is really the main advantage of EVPN as an alternative to VXLAN. If not for this, VXLAN would be simpler and preferable. Nov 2, 2020 at 19:53
  • @rnxrx, let's suppose that we have hundreds of VRFs, so option A is not an preferable choice.
    – Anton M.
    Nov 2, 2020 at 19:56


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.