Anycast addresses, a special type of Internet Protocol (IP) address, were invented in the early 1990s to simplify the process of finding replicated services (i.e., services that are provided by multiple and identical servers).1 Some of the operators of root name servers have implemented anycast addressing as a way to facilitate load sharing, to improve service, and to reduce vulnerability to attacks.
The use of anycast addresses allows a root name server operator to install copies of the root zone file at different servers (in this report, those servers that replicate the root zone file are called satellites). Properly configured and located, each of the satellites will get a share of the traffic for the root name server. Although the shares will, in most cases, not be equal, the load of queries will be distributed and thus relieve the load burden on the root name server. Satellites that are located at the same physical site are using local anycast addressing, also known as load balancing, which is widely deployed among the root name server operators.
From the user’s perspective, the great advantage derived from the adoption of anycast addressing is improved service. The satellites are typically placed at topologically diverse locations in the Internet. Queries can therefore be answered more swiftly. An additional benefit is that the DNS queries use, in the aggregate, fewer network resources, because servers will tend to be “closer” on the network to the sources of the queries.
The use of anycast addressing can sharply reduce the impact of an attack on a root name server: In the short run, physically disabling a root name server does not affect the operation of its satellites, and physically disabling a satellite disables only that satellite. In the long run, there is the question of how satellites would obtain updated root zone information. It is also much harder to mount an effective electronic attack—because queries are routed to the closest satellite (or the root name server itself, if it is the closest). An attacker would need to place (or acquire) machines close to
The ISP’s resolver then queries one of the nas.edu name servers, which refers it to a cstb.nas.edu name server, which is authoritative for the requested domain name and replies with the corresponding IP address.
A name server can answer many queries quickly when these queries request the address of a domain name for which the name server is authoritative. This is often the case, for example, for name servers on organizational intranets, where most of the requests are for IP addresses of other computers on the intranet. In such a case, the name server can respond to the query without going to the larger DNS, simply by looking up the an-