Implementation and Empirical Evaluation of Server Selection Policies for Accessing Replicated Web Services

This paper describes our work on the implementation and empirical evaluation of a client-side server selection framework for accessing replicated web services. The framework, which incorporates five representative server selection policies, was evaluated in the context of accessing a real-world replicated service, with four replicas distributed over three different continents. Our empirical results show that, in addition to the individual performance of each server hosting a replica of the replicated service, service response time using the five policies can be strongly affected by local client characteristics, such as connection capacity and workload distribution throughout the day.