You are adding EVPN VXLAN to an L3 spine and leaf topology that uses OSPF for the underlay. You plan to use a single BGP AS. Which design for EVPN neighbors is correct?

In a spine-and-leaf fabric using a single BGP AS for the EVPN control plane, the standard and most scalable design utilizes route reflectors (RRs) to avoid a full mesh of iBGP sessions between all leaf switches (VTEPs). The spine switches are the ideal location for the RR function due to their central position in the topology, connecting to all leaf switches. Leaf switches are configured as RR clients and establish iBGP sessions with the spine RRs. These sessions must be established between loopback interfaces to ensure control plane stability, as loopback interfaces remain active even if a specific physical link fails. The underlay protocol (OSPF) provides the necessary reachability between these loopback interfaces.

A. A full iBGP mesh between all leaf switches does not scale as the number of leaf switches grows, leading to excessive configuration and resource consumption. Route reflectors are the standard solution to this problem.
B. Placing the route reflector function on border leaf switches is not a recommended design. It creates an asymmetric control plane and can introduce bottlenecks, whereas spines provide a natural, centralized point for route reflection.
D. Using physical P2P link IP addresses for BGP peering is not resilient. If the specific physical link fails, the BGP session drops, disrupting the overlay, even if alternate paths exist through the fabric.

1. ArubaOS-CX 10.10 VXLAN EVPN Guide, Page 13, "EVPN configuration overview": "In a spine-leaf topology, the spine switches are configured as iBGP route reflectors, and the leaf switches are configured as route-reflector clients. This configuration avoids the requirement of a full mesh of iBGP sessions between all VTEPs." This directly supports the spine-as-RR model in option C and refutes the full-mesh model in option A.

2. ArubaOS-CX 10.10 VXLAN EVPN Guide, Page 13, "EVPN configuration overview": "BGP peering is established between the loopback interface IP addresses of the VTEPs. The underlay routing protocol provides reachability between the loopback interfaces of the VTEPs." This confirms the use of loopback IPs for peering (as in option C) and refutes the use of P2P link IPs (as in option D).

3. Aruba Validated Solution Guide: Data Center Networking with ArubaOS-CX, Page 21, "Control Plane": "The recommended best practice is to configure the spine switches as BGP route reflectors (RRs) and the leaf switches as RR clients. This simplifies configuration and scales better than a full mesh of iBGP peerings between all leaf switches." This reinforces that spines are the correct location for RRs, invalidating option B.