a root name server

If the DNS namespace is a tree, the root name servers are the receptionists standing at the very top. They do not know the address of www.example.com themselves — but they know exactly which desk to send you to next. Ask a root server about any name ending in .com, and it replies, in effect: I don't keep .com here, but the .com servers do, and here is how to reach them. They are the well-known starting point that every fresh lookup can fall back to.

Concretely, a root server is an authoritative name server for the root zone. The root zone is tiny: it mostly just lists, for each top-level domain, the name servers responsible for it. So a query to a root server for www.example.com gets a referral — the addresses of the .com name servers — not a final answer. There are 13 root server identities, labelled A through M, but that does not mean only 13 physical machines: each identity is served by hundreds of physical servers spread across the world using a trick called anycast, where many machines share the same IP address and the network simply routes you to the nearest one.

Why it matters: the roots are the anchor of the whole hierarchy and a famous example of how to make a single logical service massively robust and fast. Because the root data is tiny and rarely changes, it is cached everywhere, so in practice resolvers almost never need to bother a real root server — but the roots must still be rock-solid, since a fresh resolver with an empty cache starts there. A common misconception is that the roots see most Internet traffic; thanks to caching, they see only a thin trickle of the queries that miss every cache below them.