Approach for generating performance models from UML models of SOA systems

Model-Driven SOA is an emerging approach for developing service-oriented systems using models at different levels of abstractions and applying model transformations to generate either code or other models for the analysis of non-functional properties, such as performance. The paper proposes an approach for deriving layered queueing performance models for the evaluation of the runtime performance characteristics of such systems in the early development phases, before the entire system is built and can be deployed and measured. Early performance evaluation helps to choose an appropriate architecture, design and configuration alternatives, so that the final system meets its performance requirements. The starting point for derivation is a platform independent UML model of a SOA system representing the workflows, architecture of the underlying components offering services, and behavior of the corresponding runtime scenarios. A platform dependent model, obtained by weaving platform services into the platform-independent model through aspect-oriented modeling techniques, represents the source model for the transformation into a performance model. The deployment of the software on hardware resources is also part of the source model. The UML model is annotated with performance information by using the standard UML profile MARTE. The proposed approach is illustrated with a healthcare application