BGP multihoming: staying online when an operator goes down

The question isn't whether a link or an operator will fail one day, but when. For infrastructure whose availability matters, depending on a single provider is a risky bet. The answer is multihoming: connecting your network to several operators and letting BGP handle redundancy.

The principle

With multihoming, you announce your prefixes via at least two operators. If one becomes unavailable, BGP automatically withdraws the corresponding routes and traffic fails over to the other — with no IP address change, within moments. It's the most robust form of network high availability, because it protects against the failure of an entire provider, not just a cable.

What it takes

• An AS number and a block of addresses of your own (from your RIR).
• At least two operators offering BGP transit — ideally disjoint (different paths, POPs).
• A BGP-capable router and a controlled announcement policy.

Active/passive or load balancing?

You can favour one operator (the other as backup) or spread the traffic across both. The classic levers: local-preference for outbound traffic, AS-path prepending and BGP communities to influence inbound traffic. Inbound balancing is the trickiest — the Internet partly decides for you — but remains largely steerable with the right tools.

The pitfalls to avoid

The classic risk of a misconfigured multihoming is unintentionally becoming a transit between your two operators (their traffic flowing through you). Rigorous announcement filtering prevents it. Also mind the consistency of your RPKI ROAs so your prefixes stay valid whichever operator announces them.

Getting support

Multihoming brings maximum resilience but demands method. A solid operator as a second (or first) link makes the difference: at Connect-IX, our IP transit provisions quickly (including over GRE tunnels), with BGP communities to steer your announcements. And our managed IT can design and operate your multihomed architecture end to end.