A simulation analysis of dynamic server selection algorithms for replicated Web services

A practical approach to the provision of responsive Web services is based on introducing redundancy in the service implementation by replicating the service across a number of servers geographically distributed over the Internet. In this paper we compare the user perceived performance of three dynamic server selection algorithms operating at the client side, in order to provide some guidelines for adopting an algorithm that is appropriate for a given scenario. All the three algorithms use small probes to assess the network congestion and the server load before making a decision. The first algorithm downloads the whole requested Web page from a single server, while the other two algorithms concurrently download different pieces of the same requested Web page from different servers. The analysis is conducted via simulation with AEMPA/TwoTowers.