No abstract

Replicating web services at geographically distributed servers can offer client applications with a number of benefits, including higher service availability and improved response time. However, selecting the "best" server to invoke at the client side is not a trivial task, as this decision needs to account for (and is affected by) a number of factors, such as local connection capacity, external network conditions and servers workload. This paper presents the results of an experiment in which we implemented and empirically evaluated the performance of five server selection policies for accessing replicated web services. The experiment involved two client stations, with different connection capacities, continuously applying the five policies to invoke a real-world web service replicated over four servers in three continents. Our results show that, in addition to the individual performance of each server, service response time using the five policies is affected mainly by client differences in terms of connection capacity and workload distribution throughout the day.