Towards Automatic Construction of Reusable Prediction Models for Component-Based Performance Engineering

Abstract

Performance predictions for software architectures can reveal performance bottlenecks and quantitatively support design decisions for different architectural alternatives. As software architects aim at reusing existing software components, their performance properties should be included into performance predictions without the need for manual modelling. However, most prediction approaches do not include au- tomated support for modelling implemented components. Therefore, we propose a new reverse engineering approach, which generates Palladio performance models from Java code. In this paper, we focus on the static analysis of Java code, which we have implemented as an Eclipse plugin called Java2PCM. We evaluated our approach on a larger component-based software architecture, and show that a similar prediction accuracy can be achieved with generated models compared to completely manually specified ones.