Archivenetworking
Anycast Is Routing, Not Load Balancing
The same IP announced from many places gets users to the nearest one — and makes failures global.
- Published
- Reading time
- 1 min read
Anycast assigns one IP address to many nodes. BGP routes each client to a topologically near announcement. It is how public DNS resolvers and CDNs get low latency without GeoDNS.
The good
- Automatic latency optimization as routing changes.
- DDoS absorption: attack traffic spreads across sites.
- No client-side configuration; the address is stable.
The sharp edges
- Routing is not sticky: a BGP change can move a client mid-session. TCP connections break unless the protocol handles migration.
- Local failure is global: withdrawing a route shifts load to the next site, which may not have capacity.
- Troubleshooting is hard:
tracerouteshows one path; different clients see different backends.
Where it fits
Stateless or session-resumable services: DNS, TLS termination with QUIC migration, CDN edges. Not a primary database endpoint.
Takeaway: anycast gives you proximity and DDoS spread, but you must design for route changes and capacity shifts.