Tool Support for Performance Modeling and Optimization

Today, a great shortcoming of most of the available modeling and simulation tools for performance analysis is that model optimization is not sufficiently supported. One reason for this unsatisfying situation is the lack of qualified optimization strategies. Another reason is that modeling and simulation tools usually have a monolithic software design, which is difficult to extend with experimentation functionality. Such functionality has gained great importance in recent years because it allows to automatically extract valuable information and knowledge out of complex models. One of the most important experimentation goals is to find model parameter settings, which cause optimal model behaviour. In this paper we elaborate on the design of a powerful optimization component and its integration into existing modeling and simulation tools. For that purpose we propose a hybrid integration approach being a combination of loose document-based and tight invocation-based integration concepts. Beside the integration concept for the optimization component we also give a detailed insight into the applied optimization strategies.