Method and apparatus for load balancing of network services

A network server system for processing client requests received over a communication network includes a server farm having a number of servers. A load management switch or other network element is coupled between a set of clients and the server farm, and processes client requests for network services so as to implement a desired load balancing policy. The switch collects server statistics regarding, e.g., type of connections being serviced, delivered throughput, response time and number of active connections for each of the servers. The value of an objective function is then determined for each of the servers based on the collected statistics, and a particular one of the servers is selected for servicing one or more client requests based on the computed values of the objective function. Each of the servers may have associated therewith a unique Internet Protocol (IP) address and a common shared IP address corresponding to an advertised address of the server farm. The switch may be configured such that packets corresponding to the client requests are directed to particular servers without any modification of the IP addresses in the packets. For example, the packets may be routed by altering a local network address of a packet, such as an Ethernet media access control (MAC) address of a packet, so as to allow the packet to be delivered to a selected one of the servers.