Studying the Performance Impact of SOA Design Patterns via Coupled Model Transformations

Early performance analysis of designs for Service Oriented Architecture (SOA) can be based on performance models derived from the design models using known techniques, such as Performance from Unified Model Analysis (PUMA). When a SOA design pattern is applied to solve some architectural, design or implementation problem, it impacts the design model and its derived performance model. Conventionally, the performance model needs to be reconstructed to reflect the design pattern changes on the design model. This thesis proposes a technique to trace the causality from the design changes introduced by the pattern application to the corresponding changes in the performance model. The approach takes as input a SOA design model expressed in UML extended with two standard profiles: SoaML for expressing SOA solutions and MARTE for performance annotations. The SOA design patterns are specified using Role Based Modeling (RBM) and the performance model is expressed in Layered Queueing Networks (LQN). To support the exploration of different patterns, the thesis proposes the following approaches: 1) Systematic identification of SOA design problem, selecting an appropriate pattern and binding the design with the RBM problem specification of the pattern; 2) Systematic recording of the SOA design changes (refactoring) using the RBM pattern solution specification; 3) Automatic derivation of the corresponding performance model changes from the design model changes using coupled transformation; 4) Automatic derivation of transformation directives from the performance model changes and annotation of the performance model with the transformation directives; 5) Automatic refactoring of the performance model by QVT model transformation. Systematic and automated pattern exploration techniques and the tools support developed in the thesis are illustrated and evaluated with a Browsing and Shopping SOA case study. A test suite was designed and used to verify all the major functionalities of the proposed approach. Furthermore, several design patterns are applied to the Browsing and Shopping SOA to validate their effectiveness in the process of performance analysis by a system designer.