The DNS namespace is a tree, and every tree has a root. The root name servers are the servers that hold the root zone — the data describing which name servers are authoritative for each top-level domain. When a resolver has no cached information and must look up a name from scratch, it begins at the root. RFC 1034, “Domain Names - Concepts and Facilities,” describes this top-down resolution: a query that cannot be answered locally is referred down the tree, and the root is where that walk starts, handing the resolver onward to the appropriate top-level domain server.
The root zone is deliberately tiny and stable. It does not contain the address of every website; it contains only the pointers to the next level down. A typical lookup for www.example.com asks the root “who handles com?”, is referred to the com servers, asks them “who handles example.com?”, and is referred again to the authoritative server for that domain. Because of caching, resolvers rarely need to actually contact the root, but the root remains the ultimate point of reference for the whole system.
There are thirteen named root server identities, labeled by the letters A through M. IANA’s root server documentation describes them as the authoritative name servers that “serve the DNS root zone, often known simply as the ‘root servers’,” and notes that they form “a network of hundreds of servers in many countries around the world.” The thirteen letters are operated by twelve independent organizations — including Verisign (A and J), USC’s Information Sciences Institute (B), the University of Maryland (D), NASA (E), the Internet Systems Consortium (F), ICANN (L), and the WIDE Project (M), among others.
The reason there are only thirteen named identities is historical and technical: it kept the list of root server addresses small enough to fit in a single DNS response under the early protocol’s size limits. Each of those thirteen identities, however, is no longer a single machine. Through a routing technique called anycast, each letter is served by many physical servers spread across the globe that all answer at the same address, so that queries are handled close to where they originate. This is what gives the root its resilience: the apex of the internet’s naming tree is, in practice, hundreds of coordinated servers presenting thirteen faces to the world.