Empirical Validation and Comparison of the Model-Driven Performance Prediction Techniques of CB-SPE and Palladio

For the design of component based systems, it is important to guarantee non-functional attributes before actually composing a system. Performance usually is a crucial property of a software system, for safety or usability reasons. Several approaches to predict performance characteristics of a component based system in an early stage of development have been introduced in recent times. However, for an engineering discipline, not only the propose of techniques is needed, but also empirical studies of their applicability. This work empirically compares and evaluates two approaches to early predict the performance of component based software systems. The empirical study is conducted in form of a case study, although attempts are made to achieve a good generalizability. The results attest the CB-SPE technique a good applicability, although some problems occured. The Palladio technique has less good results. Here, there have been problems with the specification of the distribution functions.