BGP EVPN VxLAN in its first release did not provide link redundancy from the leaf switches towards the end-hosts. The traditional methodology to provide end-host link redundancy is to implement MC-LAG on the leaf switches and deploy LAGs (link aggregation) bundles on the downstream links, that is, connections towards the end-hosts.
Implementing MC-LAG or a similar networking feature requires a minimum of two switches configured with MC-LAG or similar that act as a single virtual switch to the end-hosts with redundant links. At the same, the maximum number of redundant switches that can be deployed with MC-LAG or similar is limited to two switches.
With EVPN multihoming, or EVPN-MH, a single end-host can connect to four distinct leaf or access switches where these leaf or access switches do not need to interconnect with each other to provide link redundancy.
EVPN MH leverages the flexibility and scalability of EVPN to deliver maximum uptime while staying fully integrated with its parent EVPN stack.
The following figure shows three separate hosts, each of them with separate physical connections but virtually bundled providing end-to-end connection across the EVPN fabric.
Deployment best practices
Like all scalable high-performance fabrics where virtualization and multitenancy are key requirements, the following best practices are recommended:
- EVPN-MH should be deployed whenever the fabric is based on EVPN VXLAN.
- EVPN-MH and MC-LAG cannot be deployed on the same switch.
- EVPN-MH can be deployed across four separate switches maximum.
- Current EVPN-MH implementation is for a single site only. No cross-site support is currently available.
- VRRP over EVPN-MH is not recommended. Use static anycast gateway instead.
- IGMP snooping is not supported on EVPN MH interfaces.