Anycast
Anycast means announcing the same address from several places and letting routing pick the nearest. It is the right answer for a narrow set of problems.
How it works
The same prefix is announced by BGP from multiple locations. Each network on the internet picks whichever announcement looks best from where it sits, so a user is carried to a nearby instance without any DNS trickery or redirection.
It gives you lower latency, natural load distribution and genuine resilience: if one location stops announcing, routing converges on the others within seconds.
Where it shines
- Authoritative DNS. Short, stateless, UDP. This is why nearly every large DNS operator is anycast — see running your own DNS.
- NTP and similar stateless services.
- Absorbing attacks, because a volumetric attack is split across locations rather than landing in one place.
Where it is awkward
Anything with state. Routing can change mid-session, and your TCP connection then arrives at a different machine that knows nothing about it — the connection dies. In practice routes are stable enough that short connections survive; long-lived ones, uploads and WebSockets are a real risk.
And every instance must behave identically. Anycast gives you no control over which one a user reaches, so a stale deployment in one location produces a bug that only some users see and that you cannot reproduce.
What it requires
- Your own address space, a
/24or larger for IPv4 — smaller prefixes are filtered across the internet. See bring your own IP space. - Your own AS number.
- BGP sessions at every location, with valid route objects and RPKI ROAs. See running BGP with us.
- Health-based withdrawal: when a local instance fails, that location must stop announcing, or you are routing users into a black hole. This is the part people underestimate.
The honest alternative
For most businesses, one well-connected location plus DNS-based distribution does almost as much for a fraction of the complexity. From Amsterdam you are within 10–35 ms of most of Europe already — see why Amsterdam — and load balancing covers spreading traffic without touching BGP.
Anycast is worth it when you are operating at a scale where those milliseconds and that resilience are the product. Below that scale it is a lot of machinery to maintain for a benefit nobody measures.
If you want to do it here
Talk to us. Amsterdam makes a good anycast node because of the carrier and exchange density, and we can run the session from our side. It starts as a conversation between engineers, not a form.
Still stuck? Mail support@novogara.com — an engineer answers, at any hour. Back to the knowledge base