Skip to Main Content
This paper examines the fundamental limitation of adaptation based methods in the presence of unpredictability and instability in wireless network bandwidth, server loads, and usage patterns. We argue that existing adaptation based methods may fail to produce good application response time given such unpredictability and instability, because they require somehow accurate prediction on resource conditions and usage patterns in order to perform effectively. We have designed and implemented a new, simple yet powerful, replicated client-server model which overcomes this fundamental problem of unpredictability and instability. The basic idea behind replicated client-server model is simple - application execution is replicated on both client and server, and the faster result is returned to the user. It provides the benefit of faster response time at the cost of computational overhead in replicating executions on both client and server.