Simulation-Based Performance Modeling of UML Software Architectures

Quantitative analysis of software systems is being recognized as an important issue in the software development process. Performance analysis can help to address quantitative system analysis from the early stages of the software development life cycle, e.g, to compare design alternatives or to identify system bottlenecks. Early identification of performance problems is desirable as the cost of design change increases with the later phases of the software development cycle. This thesis addresses the problem of performance analysis of software systems described at a high level of detail. We adopt a model-based approach: starting from a software model, we derive a performance model which is then evaluated. This kind of approach has the advantage of being applicable since the early software development phases; in contrast, a measurement-based approach consisting on identifying problems by direct measurements on a running system can not. We consider software descriptions as a set of annotated Unified Modeling Language (UML) diagrams. UML is a widely used notation for describing and specifying software artifacts, and recently it is being also considered for performance evaluation purposes. We define the performance model as a process-oriented simulation model. Simulation is a powerful modeling technique which can represent general and unconstrained system models, so that the software model can be more accurately represented. An algorithm for translating UML software specifications into simulation models is described. The proposed technique defines a set of annotation of UML specifications to add quantitative, performance-oriented informations. The profile is based on the UML profile for Schedulability, Performance and Time specification. The system is described in term of Use Case, Activity and Deployment diagrams. Use Case diagrams correspond to workloads applied to the system. Activity diagrams provide a high-level description of the computation steps performed by the system, and Deployment diagrams describe the physical resources on which the computations take place. A process-oriented simulation model can then be automatically derived from the annotated specification. Execution of the simulation program provides performance results that can be directly interpreted at the UML software specification level. The described algorithm has been implemented in a prototype tool called UML-Ψ (UML Performance SImulator), which is demonstrated on a case study to show the validity of the approach. The UML-Ψ tool is written in C++ and is based on a general-purpose process-oriented simulation library. It parses annotated UML models, automatically builds the corresponding simulation model and executes it. Performance results are inserted into the original UML model as tagged values, in order to give feedback to the user.