Data driven reconfiguration for performance improvements: a model based approach
Marco J. Castaldi · 2004
Self-adaptation of applications based on runtime monitoring and dynamic reconfiguration is currently considered a useful technique for managing QoS in complex software systems. In this context, many frameworks for dynamic reconfiguration have been proposed, which modify the application configuration when the threshold of a critical QoS index is crossed. In this paper we address the problem of choosing a configuration to improve the global performance of the managed application. We believe that this choice has to be driven on the basis of the results provided by online evaluation of software system models, suitably initialized with data dynamically collected in the runtime environment. We point out the issues emerged in implementing our approach, and we discuss also the characteristics that the performance models should have to be profitably used for our purposes. Finally, we describe the framework for dynamic reconfiguration in which our approach is implemented.