Glossary

EVPN Multihoming

What is EVPN Multihoming

EVPN Multihoming solves the same basic problem as MC-LAG — connecting a host to two or more switches for redundancy and load balancing — but does it natively within a BGP EVPN control plane, rather than as a separate mechanism layered on top of one. When a host is dual-homed (or multi-homed) to multiple leaf switches running EVPN-VXLAN, EVPN Multihoming coordinates those leaves so they present a single, consistent aggregation point to the host, providing all-active redundancy: every link stays actively forwarding traffic, with automatic load balancing and failover if one leaf or link goes down.

Where MC-LAG typically requires a dedicated pair of peer switches connected by a peer-link, EVPN Multihoming works directly through the EVPN control plane already carrying VXLAN routing information — meaning a host's redundant links don't need to land on the same two devices every time, and the coordination between leaves rides on infrastructure that's already there for VXLAN.

How EVPN Multihoming Works

The foundation of EVPN Multihoming is the ESI (Ethernet Segment Identifier) — a 10-byte value configured on a downlink interface. When two (or more) leaf switches configure the same ESI on their respective downlink ports connecting to the same host, those ports are recognized as peers of the same Ethernet Segment, and EVPN Multihoming treats them as a coordinated redundancy group.

Because multiple leaves can all be attached to the same Ethernet Segment, something has to decide which leaf actually forwards flooded traffic — broadcast, unknown-unicast, and multicast (BUM) traffic — arriving from the VXLAN overlay down to that segment. That's the job of the DF (Designated Forwarder): exactly one leaf per Ethernet Segment is elected DF, and only the DF is allowed to forward BUM traffic down to the local ESI; every other leaf on that segment is a Non-DF and stays silent for that traffic. DF election compares a configurable DF preference value across leaves on the same ES — the leaf with the highest preference wins, with ties broken by the lowest VTEP IP address — and the result is propagated using a dedicated Type-4 route. This is what solves what would otherwise be a serious problem: without a single elected forwarder, every leaf attached to the segment would flood the same BUM traffic down to the host redundantly.

A second mechanism, split-horizon, prevents a different failure mode: BUM traffic loops. Split-horizon enforces that only BUM traffic originating from a remote site is forwarded back down to a local Ethernet Segment — a leaf never re-floods traffic back out onto the same segment it just received equivalent traffic from — preventing both loops and duplicate delivery of the same traffic to a multihomed host.

Why EVPN Multihoming is Beneficial

  • True all-active redundancy, not active/standby: Every multihomed link stays forwarding traffic simultaneously, rather than sitting idle as a backup path the way some redundancy schemes do.

  • Scales past a fixed device pair: Because coordination happens through the shared ESI concept rather than a dedicated peer relationship between exactly two switches, a host's redundant links aren't locked into always landing on the same pair of leaves.

  • Solves BUM replication cleanly: DF election guarantees exactly one leaf forwards flooded traffic to a given segment, avoiding the duplicate-delivery problem that would occur if every attached leaf flooded independently.

  • Loop-free by design: Split-horizon logic prevents BUM traffic from looping back out the segment it arrived from, without requiring a separate spanning-tree-style mechanism to catch it.

  • Built on infrastructure the network already has: Because EVPN Multihoming rides on the same BGP EVPN control plane used for VXLAN routing, it doesn't require standing up a parallel redundancy protocol alongside it.

At Asteraix

What We Can Do at Asteraix

AsterNOS implements EVPN Multihoming as a native extension of its EVPN-VXLAN stack, configurable through the CLI alongside cross-device link aggregation.

  • Two ways to configure ESI: A 10-byte ESI can be set directly with evpn mh es-id <10-byte-value>, or built from a shorter es-id (1–16777215) combined with evpn mh es-sys-mac <mac> — giving operators either full manual control or a simpler, MAC-derived identifier depending on what fits their addressing scheme.

  • Configurable DF preference: evpn mh es-df-pref <preference> (range 1–65535, default 32767) lets operators directly influence which leaf wins Designated Forwarder election for a given Ethernet Segment, rather than leaving the outcome entirely to VTEP IP tiebreaking.

  • Tunable hold timers for real deployments: evpn mh mac-holdtime and evpn mh neigh-holdtime (both 0–86400 seconds, with AsterNOS's own documentation recommending 300 and 360 seconds respectively) control how long synchronized MAC and neighbor entries stay valid after the learning switch's own entries age out — tuned for real failover behavior rather than left at a generic default.

  • Redirect-off for faster downlink convergence: evpn mh redirect-off, used in AsterNOS's own documented deployment example, ensures fault convergence happens correctly on the downlink side of a multihomed connection.

  • Unique-IP support for Layer 3 access-side peering: evpn unique-ip vlan <vlan-id> paired with evpn mh disable-advertise-svi-mac lets a multihomed pair run individual BGP sessions with an access-side device over a shared-gateway VLAN — demonstrated in AsterNOS's reference topology establishing real eBGP peering with BFD to a downstream ToR switch.

  • Full visibility into ES state: show evpn es (with detail or json output) shows every configured Ethernet Segment's type — Local, Remote, Bypass, or Non-DF — directly from the CLI, alongside show vxlan tunnel and show link-aggregation summary for confirming the underlying VXLAN and LACP state a multihoming deployment depends on.