If your visitors are spread across the globe, the distance their requests travel directly affects how fast your service feels. Anycast is a clever networking technique that sends each user to the nearest copy of your service automatically, and it underpins much of the fast, resilient internet we rely on.
What Anycast Is
Anycast is a network addressing method where the same IP address is announced from many locations at once. When a user connects, the network routes them to whichever location is closest in network terms. One address, many servers, and the internet quietly picks the best one for each visitor.
How It Differs from Unicast
Traditional unicast gives every server a unique address, so a request always goes to one specific machine no matter where the user is. Anycast shares a single address across many servers, letting the network choose the nearest. This is the difference between everyone calling the same distant office and everyone reaching their local branch.
Why Anycast Speeds Things Up
Shorter network paths mean lower latency, and lower latency means faster responses. By routing users to a nearby node rather than a single far-off server, anycast cuts round-trip times significantly. For latency-sensitive services like DNS and content delivery, this makes a noticeable difference.
Built-In Resilience
Anycast does more than improve speed. If one location goes offline, the network simply stops announcing the address from there and routes users to the next closest node. This automatic failover makes anycast naturally resistant to outages and a strong defense against traffic floods.
Where Anycast Is Used
Anycast powers many of the internet's most critical services. The DNS root servers, large public DNS resolvers, and content delivery networks all rely on it to stay fast and available worldwide. It is also a key tool in absorbing and spreading out denial-of-service traffic across many locations.
Things to Keep in Mind
Anycast works best for stateless or short-lived requests, since a long connection could theoretically be re-routed mid-session. For most web, DNS, and CDN traffic this is rarely an issue, but it is why anycast is favored for those use cases rather than for long stateful sessions.
Anycast vs Unicast: When Each One Is the Right Answer
Unicast is the default: one IP address, one server, one place, and BGP delivers every packet to that single destination. Anycast announces the same prefix from many locations at once, and the routing system delivers each packet to whichever announcement is "closest" by BGP's rules, usually the shortest AS path. If one location withdraws its announcement, traffic shifts to the next-closest within seconds, with no DNS change and no failover script.
| Unicast | Anycast | |
|---|---|---|
| Failover speed | DNS TTL bound, often minutes | BGP convergence, roughly 10–30 seconds |
| Works for | Any protocol, stateful or not | Short, stateless exchanges: DNS, NTP, HTTP with edge termination |
| Control over who lands where | Exact | Whatever BGP decides; not always the geographically nearest PoP |
| Requirements | One IP from any provider | Your own ASN, provider-independent space (a /24 IPv4 or /48 IPv6 minimum), RPKI ROAs, BGP at every PoP |
| DDoS behaviour | The whole attack lands on one server | Spread across every PoP announcing the prefix |
Where anycast struggles
Long-lived TCP sessions can be moved mid-flight to a different PoP when BGP paths change, and the new PoP has no record of the connection. Production deployments handle this by keeping sessions short, terminating TLS at the anycast edge and using a unicast connection to a fixed origin, or replicating state between PoPs. Two more traps: BGP does not know a PoP is saturated and will keep sending traffic until you withdraw the route, so health-driven announcements are mandatory; and debugging needs every log line tagged with the PoP that served it.
Anycast vs GeoDNS
GeoDNS solves a similar problem in a different layer: it hands different IPs to users in different regions. It works for any protocol and gives exact control, but its failover is bound to DNS TTLs. Many operators combine the two, GeoDNS to pick a region and anycast inside each region for instant failover.
Three deployment patterns that work
- Pure anycast, stateless. DNS, NTP, simple HTTPS APIs. Every PoP serves identical content and failover is automatic.
- Anycast edge, unicast origin. TLS termination, caching and WAF at the anycast edge; the edge talks to an origin over unicast. This is how every CDN is built.
- Anycast front door, application routing. The nearest PoP accepts the request, inspects it, and forwards to the right region. Common in enterprise SaaS.
To run your own anycast you need multiple data centers with BGP-capable upstreams and identical services at each one. On AS60982 that means announcing your prefix from the NODED PoPs you pick and letting BGP do the routing; see the BYOIP and BGP page for the prefix requirements.
How Noded Can Help
If you serve a global audience and want lower latency and better resilience, we can help you take advantage of anycast routing. Tell us where your users are and what you are serving, and we will help you design a network that reaches everyone quickly.
FAQ
Is anycast the same as a CDN?
Not quite. A CDN often uses anycast to route users to nearby edge servers, but anycast is the underlying routing technique while a CDN is the broader content delivery service built on top of it.
Does anycast improve reliability?
Yes. Because the same address is announced from many locations, the network automatically routes around any node that fails. This built-in failover makes services more resilient to outages and attacks.
Can I use anycast for any application?
Anycast is best suited to stateless or short requests like DNS and content delivery. Long stateful connections can be trickier because routing changes could move a session, so those use cases need extra care.
Why is anycast good against DDoS attacks?
Anycast spreads incoming traffic across many locations instead of concentrating it on one server. This dilutes attack traffic and lets each node absorb a smaller share, improving your ability to stay online.