Performance prediction for software architectures

The quantitative evaluation of certain quality attributes— performance, timeliness, and reliability— is important for component-based embedded systems. We propose an approach for the performance estimation of component-based software that forms a product family. The proposed approach, Analysis and Prediction of Performance for Evolving Architectures (APPEAR), employs both structural and stochastic modeling techniques. The former are used to reason about the properties of components, while the latter allow one to abstract from irrelevant details of the execution architecture. The method consists of two main parts: (1) calibrating a statistical regression model by measuring existing applications and (2) using the calibrated model to predict the performance of new applications. Both parts are based on a model of the application in order to describe relevant execution properties in terms of a socalled signature. A predictor that is determined by statistical regression techniques is used to relate the values of the signature to the observed or predicted performance measures. APPEAR supports the flexible choice of software parts that need structural modeling and ones that statistical modeling. Thereby it is assumed that the latter are not seriously modified during the software evolution. The suggested approach is being validated with two industrial case studies in the Consumer Electronics and Professional Systems domain.