Despite the increasing computational power of modern computers, many large, distributed software systems still suffer from performance problems today. To avoid design-related performance problems, modeldriven performance prediction methods analyse the response times, throughputs, and resource utilisations of systems under development based on design documents before and during implementation. For component-based software systems, existing prediction methods neglect the performance influence of different usage profiles (i.e., the number of requests and the included parameter values) in their specification languages, which limits their prediction accuracy. This thesis proposes new modelling languages and according model transformations, which allow a reusable description of usage profile dependencies in component-based software systems. The thesis includes an experimental evaluation, which shows that predictions based on the newly introduced models can support design decisions for scenarios, whose performance is influences by different usage profiles.
Mendeley saves you time finding and organizing research
Choose a citation style from the tabs below