Many large web sites get more than 100 million hits everyday. They need a scalable web server system that can provide better performance to all the clients that may be in different geographical regions. Higher delays and losses are common on WAN links. To provide a better service to all the clients, it is natural to have fully replicated web server clusters in different geographical regions. In such an environment, one of the most important issue is that of server selection (and load balancing). The client's request should be directed to one of the servers in a way that the response can be quick. We assume that web servers are functionally homogeneous, i.e. any one of them can serve any client request. Another important point is that this system should not require modification of any client side component or existing standard protocol.
In this thesis, we have developed a test bed to emulate the world wide web environment and compare different schemes. A large number of systems have been proposed to do this load balancing. We also propose a new scheme which is based on estimating the round trip time between the client and various server clusters. The proposed scheme is shown (through emulation) to perform significantly better than many of the existing scheme.