What is Anycast DNS — how global edge routing eliminates latency
Executive Summary & Core Definition
Anycast routing allows multiple physical servers worldwide to share a single IP address, routing website visitors to the nearest datacenter automatically.
Anycast is a routing technique in which one IP address is shared by a group of endpoints, each of them in a different place. Microsoft Learn's own overview puts it plainly: anycast "provides multiple routing paths to a group of endpoints that are each assigned the same IP address", and "routing protocols are used to choose which is the best destination".
With Anycast DNS, a DNS server — or a group of servers — answers a query "based on the geographical location of a DNS client". The same address is announced from every location that serves it, the network picks one, and the visitor gets an answer from the closest healthy copy rather than from one machine in one datacenter.
Unicast vs. Anycast Routing
Understanding anycast requires contrasting it with traditional unicast routing:
- Unicast Routing: One IP address maps to one physical server location. If that server is in Frankfurt, Germany, a visitor in Sydney, Australia must send network packets halfway around the planet, incurring hundreds of milliseconds of routing latency.
- Anycast Routing: One IP address is announced simultaneously by nodes in Tokyo, London, New York, São Paulo, and Sydney. Every visitor is answered by a node near them.
The same address, announced from every edge, is the whole of it:
textone address: 203.0.113.10 │ ├─ announced from Tokyo ─ the visitor in Tokyo ├─ announced from London ─ the visitor in London ├─ announced from New York ─ the visitor in New York └─ announced from Sydney ─ the visitor in Sydney
Under unicast that one address would point at one place, and the visitor in Sydney would cross an ocean to reach it. BGP is what turns one address into many — and equal-cost multi-path routing is what lets the network spread the traffic between them.
How a query actually reaches the right node
This is the part that is usually told as a guarantee, and it is not one. Microsoft's description is more careful than the folklore: when a client queries an anycast address, "the available routes are evaluated, and the DNS query is sent to the preferred location. In general, this location is the closest based on network topology."
So the accurate statement is: the network sends the query to the shortest available route it can see, and that route is usually — in general — to the nearest location, but it is the network topology, not the distance on a map, that decides. A route that detours can still be the shortest one available to that client at that moment.
What it is used for
Microsoft lists three, and they are the honest three:
| What anycast DNS is for | What it actually does |
|---|---|
| Response time | A query is answered from a location near the client instead of a fixed origin, which "can enhance DNS response time" |
| Redundancy | "An extra layer of redundancy" — if one location is down or its route is cut, the address is still answered from the others |
| Denial of service | It "can help protect against DNS denial of service attacks", because a flood aimed at one address is spread across every copy of it |
And it is not an experiment. Anycast DNS "is used commonly today to route DNS traffic for many global DNS services. For example, the root DNS system depends heavily on Anycast DNS" — and Azure DNS uses anycast to serve its own zones.
What it is not
- It does not guarantee the nearest node. It guarantees that the network
picks a good one, and "good" is measured in routes, not kilometres.
- It is not required for DNS to scale. Microsoft's own FAQ is blunt:
anycast DNS "works seamlessly with an on-premises DNS service. However, Anycast isn't required for the DNS service to scale."
- It does not make the site fast by itself. It makes the lookup fast
and local. What happens after the answer arrives — how quickly the page renders — is a different metric, and the one the rest of this site measures.
Native Anycast on Zeso
The site itself is served over the same kind of network: Zeso compiles websites to immutable static assets distributed across ${EDGE_CLAIMS.locations} Anycast edge locations, and a response comes from the city closest to the visitor rather than from a single origin datacenter.
For the domain lookup itself, the nameservers are Cloudflare's — which means the answer to "where is this domain" is computed by the same anycast principle: at the location closest to the person asking, not at one machine somewhere else.
Sources
- Microsoft Learn — Anycast DNS overview — the definition, the "preferred location … in general the closest based on network topology" wording, the three uses, the root-servers dependency, and the FAQ that anycast is not required for DNS to scale. Last updated 2023-02-13.
- RFC 4786 §4.4.3 — Equal-Cost Paths, which Microsoft Learn cites for the "same back-end DNS server, consistent BGP path" case.
- Wikipedia — Anycast
Experience edge-native speed on your domain
Sub-50ms TTFB, Anycast BGP routing, and automatic Let's Encrypt certificates across 330+ locations.