A scalable domain name system with persistence and load balancing receives requests from client DNS servers or other DNS servers. Each DNS server is associated with a subset of the DNS groups in the network and the invention checks to see if the client DNS server is part of the DNS group that the DNS server is authoritative. If the DNS server is not authoritative for the client DNS server's group, then the request is forwarded to the proper DNS server. Otherwise, the invention checks a persistence table to see if a persistent response is required for the request. If a persistent response is required, the appropriate IP address entry in the table is returned to the requestor. If a persistent response is not required, the invention determines the load, availability, and latency of the content servers from information stored in a latency table to determine the proper content server's address to return to the requestor. The DNS server sends any responses to forwarded requests to the client DNS server. When more than one DNS server is authoritative for an overlapping DNS group, the invention ensures that the persistence and latency tables are synchronized with any other DNS servers. The invention also allows the persistence and latency tables to be located on a set of table servers within the network. Another aspect of the invention allows the DNS server that a request is forwarded to, to respond directly to the client and the client is redirected to the DNS server.