Completion and Extension Techniques for Enterprise Software Performance Engineering

Software performance engineering supports software architects in identifying potential performance problems in software systems during the design phase. Details of the implementation and execution environment of a system are crucial for accurate performance predictions. Yet, only little information about these details is available during early stages of the software life-cycle; furthermore, model-based architectural description languages used by software architects are lacking support for performance-relevant information. Architectural models need to be extended, so they are ready to include design details as they become available when development advances. Model extensions, however, give rise to the problem of model and metamodel evolution. We report on our experiences with a variety of metamodel extension techniques, covering completions, direct invasive techniques, decorator models, and profiles in the context of model-driven performance engineering. Our goal is to enable performance engineers to find the optimal solution when metamodel variability and evolution support is required. In a case study, we extend a component-based system with thread management information, thereby illustrating the benefit of performance completions with respect to the accuracy of performance predictions.