Summary:

• Neutron side implementation of BGP EVPN is still pending. This is a big challenge from end to end perspective. 
• Since VRFs are implemented in kernel networking, fast datapath does not have an option for EVPN right now.
• I believe BGP EVPN for the controller network is possible with manual intervention.
• There is an open question with bgp evpn on having geneve (compute to compute) and vxlan tunnel (compute to DC gateway) on the same compute node. I need to try this in the lab. So, created this EPIC (will divide in further tasks) to target some FRx.

Gurpreet's ask:

(1) What is the scope of work for supporting E2E (neutron) for BGP-EVPN?

(2) From the POC perspective, it is ok to not have neutron support and not have DPDK support. The objective of the POC is to uncover any functionality gaps and to get a performance baseline to determine if the performance will work for VMO2 or other customers who have kernel-only deployments. 

(3)  Getting an idea of ETA for OVN native BGP support together with the POC results will help us in deciding whether we should complete the neutron work and other associated gaps or should continue the productization of BGP-OVN (Native).