Coupled model transformations

Model-driven performance prediction methods use abstract design models to predict the performance of the modelled system during early development stages. However, performance is an attribute of the running system and not its model. The system contains many implementation details not part of its model but still affecting the performance at run-time. Existing approaches neglect details of the implementation due to the abstraction underlying the design model. Completion components [26] deal with this problem, however, they have to be added manually to the prediction model. In this work, we assume that the system's implementation is generated by a chain of model transformations. In this scenario, the transformation rules determine the transformation result. By analysing these transformation rules, a second transformation can be derived which automatically adds details to the prediction model according to the encoded rules. We call this transformation a coupled transformation as it is coupled to an corresponding model-to-code transformation. It uses the knowledge on the output of the model-to-code transformation to increase performance prediction accuracy. The introduced coupled transformations method is validated in a case study in which a parametrised transformation maps abstract component connectors to realisations in different RPC calls. In this study, the corresponding coupled transformation captures the RPC's details with a prediction error of less than 5%.