Dell SmartFabric OS10 User Guide Release 10.5.3

BGP peering for underlay IP routes

eBGP peering session between Leaf and Spine using direct interface IP address.

iBGP session between VLT peers on dedicated VLAN configured over VLTi.

Following illustration shows setting up of eBGP peering with VLT:

BGP peering for Overlay services

This section describes eBGP Peering between loopback IP addresses between Leaf and Spine.

The following illustration depicts BGP peering for overlay services:

Each BGP router has two separate BGP sessions between each Leaf and Spine - one for underlay routes and one for overlay routes (Virtual Network MACs).

The following sequence describes the topology depicted in the BGP peering for overlay services image:

1. This topology requires two separate loopback interfaces.
2. BGP underlay session - IPv4 address family - used to learn underlay IP route reachability to remote VTEPs.
   1. Uses the point-to-point link IP address as the BGP neighbor IP address.
   2. IP reachability to both loopback IP addresses are advertised with the point-to-point IP as Next-hop.
   3. In case of failure of all network links on one of the VLT nodes, BGP sessions between VLT peers are needed for IPv4 address family to retain underlay routes.
3. BGP Overlay Session - EVPN address family - used to exchange overlay Layer 2 information between the VTEP nodes (Leaf nodes).
   1. Uses a unique loopback interface IP on each node (Leaf and Spine) as the BGP neighbor IP address.
   2. An explicit override command (update-source) is needed to ensure that the local IP address that is used in the TCP session is the loopback IP address instead of the link IP address. This behavior ensures that the session is retained even if the links fail.
   3. At the VTEPs, irrespective of the update-source override, the EVPN routes are always advertised with the NVE source IP address (Lo1) as the Next-Hop.
   4. eBGP next-hop modification is disabled by default on all EVPN routes. So, the Spine nodes do not use their own Next-hop when advertising the EVPN routes to the remote VTEPs.
   5. There is no EVPN address family peering between the VLT peer nodes within the same VLT domain.
4. By default, Dell SmartFabric OS10 does not override the Next-hop when advertising EVPN address family routes to even eBGP peers. So, EVPN routes received by Leaf nodes contain the originating Leaf node's IP address in the Next-Hop.
5. If all Leaf nodes are in the same AS, then the Spine nodes need to be explicitly configured to disable sender-side-loop-detection. Without this configuration, the eBGP on Spine does not distribute routes received from VTEP A (AS2) back to VTEP C (also AS2).
   NOTE: You must enable loop detection on the Leaf nodes (default). This configuration ensures that the Leaf VTEP nodes do not distribute the routes back to other Spine nodes again.
6. You can use any IP address that is unique to a node as the BGP Router ID. Typically, a separate loopback interface with static IP address is used for this purpose to avoid allocating this IP address for other purposes.
7. Even when a VLT node loses all its direct network links to the fabric, it still has reachability to its previous underlay First-hops (Spine or RR) over the VLTi with the VLT peer node as the new underlay First-hop.
   1. A VLT node's BGP EVPN sessions on loopback interfaces do not terminate even if it loses all direct connectivity to the underlay network.
   2. This behavior requires the BGP multi-hop parameter to be configured as 3.

The following sequence depicts the setup for common BGP session for underlay and overlay:

1. The same BGP session advertises both IP routes and EVPN address family routes.
2. The Network link IP address on each node is used as the BGP neighbor IP address.
3. A loopback interface IP address is needed on each node to act as NVE source IP address.
4. IP reachability to the loopback IP address is advertised with the link IP address as Next-hop.
5. At the VTEP, EVPN routes are always advertised with the NVE source IP address as Next-hop.
6. The Spine is internally setup to not override the EVPN address family route next-hops with its own IP address.
7. Any IP address that is unique to a node can be used as the BGP Router ID. Typically, a separate loopback interface with static IP address is used for this purpose to avoid allocating this IP address for other purposes.