| As Web applications become more and more prevalent, the performance of Web applications has become more and more important. In a client-server architecture, the client cannot affect many of the factors that influences performance of a Web application but can choose how to interact with the server. In this thesis, we explore the Web application optimization using client side adaptation of configuration parameters. Properly optimizing the setting of configuration parameters can greatly improve performance, especially in the presence of changing server workloads. We propose approaches to dynamic throughput optimization of client systems, focusing on parameters like batch size (the number of queries that can be batched in a single call to the server) and concurrency (the number of simultaneous connections to the server). We investigate two optimizers - one based on Bayesian optimization and the second based on Fuzzy control. In all cases, results show that by tuning the parameters to the Web server's behavior, the performance of the Web application can increase dramatically. In comparison, Bayesian optimization outperforms fuzzy control in terms of average throughput that is achieved, robustness, and adaptations to changing workloads. Moreover, Bayesian optimization is generic and can simultaneously optimize multiple parameters. |
