The increasing complexity of software due to continuous technological advances has motivated the use of models in the software development process. Initially, models were mainly used as drafts to help developers understand their programs. Later they were used extensively and a new discipline called Model-Driven Engineering (MDE) was born. In the MDE paradigm, aside from the models themselves, model transformations (MT) are garnering interest as they allow the analysis and manipulation of models. Therefore, the performance, scalability and correctness of model transformations have become critical issues and thus they deserve a thorough study. Existing model transformation engines are principally based on sequential and in-memory execution strategies, and hence their capabilities to transform very large models in parallel and in distributed environments are limited. Current tools and languages are not able to cope with models that are not located in a single machine and, even worse, most of them require the model to be in a single file. Moreover, once a model transformation has been written and executed-either sequentially or in parallel-it is necessary to rely on methods, mechanisms, and tools for checking its correctness.
In this dissertation, our contribution is twofold. Firstly, we introduce a novel execution platform that permits the parallel execution of both out-place and in-place model transformations, regardless of whether the models fit into a single machine memory or not. This platform can be used as a target for high-level transformation language compilers, so that existing model transformations do not need to be rewritten in another language but only have to be executed more efficiently. Another advantage is that a developer who is familiar with an existing model transformation language does not need to learn a new one.
In addition to performance, the correctness of model transformations is an essential aspect that needs to be addressed if MTs are going to be used in realistic industrial settings. Due to the fact that the most popular model transformation languages are rule-based, i.e., the transformations written in those languages comprise rules that define how the model elements are transformed, the second contribution of this thesis is a static approach for locating faulty rules in model transformations. Current approaches able to fully prove correctness-such as model checking techniques-require an unacceptable amount of time and memory. Our approach cannot fully prove correctness but can be very useful for identifying bugs at an early development stage, quickly and cost effectively.