A method and system for load balancing over a cluster of authentication, authorization and accounting (AAA) servers. The method performs a distribution of AAA requests among AAA servers having an active AAA connection with an AAA client. The method includes establishing TCP connections with a plurality of AAA servers, using a TCP connection request received from at least one AAA client; opening AAA connections with a plurality of AAA servers, using an AAA connection request received from at least one AAA client, and distributing AAA requests to AAA servers with an active AAA connection according to a predefined load balancing algorithm. The method is further capable of multiplexing outbound messages and requests received from a plurality of AAA servers. The AAA protocol supported by the method includes, but is not limited to, a Diameter protocol, a lightweight directory access protocol (LDAP), and the likes.