Building Next-Generation Data Center

12 module online course

Start now!

Using BGP in a Data Center Leaf-and-Spine Fabric

ipSpace.net » Documents » Using BGP in a Data Center Leaf-and-Spine Fabric

Large data center leaf-and-spine fabrics often use BGP to propagate endpoint reachability information (using IPv4, IPv6 or EVPN address family), or to replace IGP.

In both cases, you have to decide whether to use IBGP or EBGP within the fabric, and whether to use one or multiple AS numbers on the leaf layer and the spine layer.

This document covers the following design decisions:

  • Should we use BGP as a routing protocol within a data center fabric?
  • Should we use IBGP or EBGP within the leaf-and-spine fabric?
  • Should we use the same AS number on all spine switches?
  • Should we use the same AS number on all leaf switches, or should each leaf switch have a unique AS number?
  • Should leaf switches in an MLAG cluster use the same AS number, or should they have a unique AS number?
  • What’s the impact of AS numbering on EVPN and MPLS/VPN services?
Extensive feedback from Dinesh Dutt and Nicola Modena made this document significantly better. Thank you!

Before You Start

The rest of the document assumes reasonable familiarity with BGP concepts (for example, the details of next-hop handling in BGP or the difference between IBGP and EBGP).

If you’re building a small data center fabric no larger than a few dozen switches, and are not familiar with BGP use these rules-of-thumb while keeping in mind you should understand recipes you use:

  • Use OSPF or IS-IS as the fabric routing protocol.
  • Use BGP only when needed, for example to run EVPN control plane.

When you plan to run BGP in combination with an underlying IGP:

  • Use IBGP within a single data center (or pod) and EBGP between them
  • Use spine switches as BGP route reflectors.
Running IBGP in combination with IGP in a simple data center fabric

More information