Model-driven performance measurement and assessment with relational traces

Model Driven Engineering is an emerging approach in software development which argues using models as primary software artifacts. The models can vary from prescriptive ones specified in Domain Specific Modeling Languages (DSMLs), for the domain specific software development (e.g. Graphical User Interfaces DSML), to descriptive models for analysis, such as queuing networks and Petri-nets. The main benefit of such approach is that it reduces the time and effort spent in implementation by transforming the domain specific prescriptive models to the code. Furthermore, performance characteristics of the implementation can be predicted during the design phase with usage of transformation of domain specific constructs to formal mathematical models. For software performance, as an important software Quality of Service attribute, several approaches for prediction exist in MDE. However, MDE still lacks a systematic approach for performance measurement and metrics assessment. This thesis presents MoDePeMART, an approach for Model Driven Performance Measurement and Assessment with Relational Traces. The approach suggests declarative specification of performance metrics in a domainspecific language and the usage of relational databases for storage and metric computation. The approach is evaluated with the implementation of a UML Profile for UML Class and State diagrams and transformations from the profile to a commercial relational database management system.