Anycast

Anycast is an addressing and routing technique in which a single IP address is advertised from multiple physical locations at once. When a client connects to that address, the network's routing fabric delivers the packets to whichever instance is closest in routing terms, not to one fixed machine. This is the opposite of unicast, where one address maps to exactly one host and all traffic lands there. The result: the same service IP can be served by dozens of geographically dispersed nodes, while every user transparently reaches the nearest one — with no special client software and no awareness that anything unusual is happening.

A concrete example makes the magic tangible. When your node or laptop queries the DNS resolver at a well-known public address, that address answers from hundreds of cities; a user in Montreal reaches a nearby instance, a user in Singapore reaches theirs, and both typed the same four numbers. Neither user chose a server, no directory redirected them, and if the Montreal site went dark mid-afternoon, its users would slide to the next-nearest instance without a settings change. Contrast that with the unicast world, where reaching a distant service means every packet crossing an ocean and a regional outage means downtime. Anycast quietly deleted an entire category of fragility from the internet's most critical services.

How BGP makes it work

Anycast is built directly on BGP, the protocol that stitches the internet's independent networks together. Each anycast location announces the identical IP prefix to its upstream providers, and BGP's best-path selection naturally steers each requester toward the announcement reachable at the lowest cost — factoring in path length, transit relationships, and local routing policy. There is no coordinator and no load-balancing appliance making these decisions; the internet's ordinary routing behavior does all the work. The failover story is the elegant part: if one node fails or withdraws its announcement, routes simply reconverge onto the remaining instances within the normal BGP convergence window. Clients need no reconfiguration, no DNS change, and usually notice nothing.

Where it runs the internet

Anycast underpins the most resilience-critical infrastructure there is. The DNS root servers are the canonical example: the thirteen root server addresses are served by well over a thousand physical instances worldwide, all via anycast — which is why no single datacenter failure, or even a regional disaster, can take root DNS down. Public resolvers and content delivery networks use the same trick to spread load and absorb distributed denial-of-service attacks: a flood aimed at one address gets diluted across every site announcing it, with each location absorbing only its regional share. Anycast suits short, stateless exchanges best — DNS queries are the perfect fit — because a mid-connection route change can move long-lived sessions to a different server; engineering around that is a known, solvable nuisance rather than a blocker.

Limits worth knowing

"Closest in routing terms" is not "closest in kilometers" — BGP optimizes for policy and path attributes, so traffic occasionally lands somewhere geographically surprising. Anycast also requires the ability to announce address space via BGP, which means owning a portable prefix and an autonomous system number or renting that capability — historically the province of network operators, though small-scale BGP has become far more accessible to determined individuals and co-ops.

The sovereignty reading

Anycast deserves attention from decentralization-minded operators because it is centralization-resistance built from pure protocol, with no company in the loop. It turns several independent points of presence into one robust, attack-resistant, location-aware service using nothing but routing. Bitcoin's own peer-to-peer design solves discovery differently — every full node is equal and none share an address — but the infrastructure your node quietly depends on, from DNS seeds to time sources, very likely reaches you over anycast. It pairs naturally with a load balancer at each site — anycast picks the nearest region, the balancer spreads traffic across servers inside it — and with a reverse proxy terminating connections at each edge.