In the EVPN Route Server mode, the tier-0 service router (SR) hosted on the edge node acts as a BGP route server, establishing BGP control plane sessions with southbound VNFs and external data center routers. ESXi hypervisors exchange the user plane traffic directly with the data center fabric routers using VXLAN encapsulation, bypassing the edge node in the data path.

From the BGP control plane perspective, there are two types of sessions:

Session Type Description
Between hosted VNFs and tier-0 VRF gateway.
  • BGP IPv4 unicast and IPv6 unicast sessions from the VNF to the tier-0 VRF service ports.
  • IP prefixes learned from the VNF via the BGP IPv4/IPv6 unicast sessions are advertised as EVPN Route-Type 5 towards the external route with the corresponding VRF route distinguisher and route targets.
  • IP prefixes (RT-5) learned from the external router via the BGP EVPN session are injected in the tier-0 VRF routing table based on the route target policies and are advertised as IPv4/IPv6 unicast routes to the VNF.
Between tier-0 SR and DC gateways.
  • BGP IPv4 session with L2VPN EVPN address family from tier-0 SR to the loopback of DC gateways.
  • The IP prefixes learned from the VNF via the BGP IPv4/IPv6 unicast sessions are advertised as EVPN Route-Type 5 towards the external router with the corresponding VRF route distinguisher and route targets.
  • IP prefixes (RT-5) learned from the external router via the BGP EVPN session are injected in the tier-0 VRF routing table based on the route target policies and are advertised as IPv4/IPv6 unicast routes to the VNF.
  • EVPN routes of type 2 (RT-2) are exchanged between tier-0 SR and DC gateways to advertise MAC and IP/MAC bindings for RT-5 routes next hops.
  • EVPN routes of type 1 (RT-1) are sent by the DC gateway to the tier-0 SRs when multiple data center gateways are configured as L2ECMP EVPN multihoming. RT-1 routes are used to create a list of data center gateway VTEPs to be used as next hop by ESXi nodes.

The NSX-T EVPN Route Server mode is based on the "Interface-ful IP-VRF-to-IP-VRF with SBD IRB" as defined in the IETF RFC 9136. The RFC 9136 introduces a concept called overlay index in EVPN. A key concept of EVPN RT-5 is the overlay index, which can be a gateway IP address, a MAC, or an ESI. When a node receives an EVPN RT-5 with an overlay index specified, the receiving node performs a recursive route resolution to find the appropriate node to forward the data packets for the corresponding IP prefix.

NSX-T EVPN Route Server mode implements the gateway IP address as the overlay index. The tier-0 SR also advertises to the external router an additional EVPN type-2 route with the appropriate MAC/IP (gateway IP) binding and the corresponding VXLAN TEP address.

The gateway IP address in this case will be the IPv4 BGP next hop for a given prefix as advertised by the VNF to the tier-0 VRF gateway. The recursive route resolution uses respective RT-2 to learn the ESXi TEP address where the VNF is hosted.

Data Center Gateway Requirements

The data center gateway router connected to the edge node tier-0 SR must support the "Interface-ful IP-VRF-to-IP-VRF with SBD IRB Mode" described in the IETF RFC 9136, section 4.4.2.

Virtual Network Function (VNF) Requirements

A VNF is typically a virtual machine used for some networking function such as a virtual router, firewall, or a Telco 5G core application. In the context of EVPN Route Server mode, the VNF is hosted by an ESXi hypervisor and should support 802.1Q-tagged interfaces and regular BGP protocol with IPv4 and IPv6 unicast address families.