| The emergence of the Model-Driven Architecture (MDA) has changed the focus of software development from code to models, and raised the need for model transformations. The objective of this thesis is to develop a transformation of a UML model extended with MARTE performance annotations in to a more abstract performance model, which can then be used for the analysis of real-time properties. Up until now, the main method of implementing this kind of transformation has been through general-purpose languages such as Java. But this approach is not entirely consistent with the goals of MDA for reasons that will be further discussed.;The goal of this thesis is to use a special-purpose model transformation language to transform UML 2.1 Sequence Diagrams (with applied MARTE stereotypes) to Core Scenario Models. This transformation represents a task from a larger project entitled Performance by Unified Model Analysis (PUMA). The transformation language employed in this study is the ATL language by the ATLAS Group. As a hybrid between declarative and imperative rules, ATL is a fairly distinct language from general-purpose object-oriented languages. However, the use of this language reveals many advantages over the alternative approach, as will be demonstrated. Be that as it may, neither ATL nor other rule-based transformation languages like MOF-QVT have been around that long, and so conventional wisdom on their use remains narrow. One of the areas further explored in this thesis is the development of design strategies and patterns for model transformations. |
