User-Centered Performance Engineering of Model Transformations

In Model-Driven Engineering, models are key artifacts. Due to the fact that the systems to be developed become larger and more complex, the corresponding models also become larger and more complex. This trend also influences operations on these models, such as transformations. They are applied at design time and at runtime, e.g. to update models, generate code or to create new models. With increasing model size, their execution time increases, making their performance an important quality aspect. Current research mainly concentrates on further improvements of the transformation engine that performs the transformation, but this will not solve the problem alone. Engine optimizations will never be able to mitigate every possible performance problem due to the fact that there's an arbitrary amount of ways to define a transformation as well as the models and meta-models that all affect the runtime. Therefore, transformation engineers must also ensure that they define their transformations in such a way that they have a short execution time. To achieve this, a performance engineering approach for model transformations is necessary. This approach must consist of steps and techniques that help to analyze and improve performance. In this paper we present our performance engineering approach for declarative model transformations. We identified the five artifacts Guidelines, Monitoring, Analyses, Visualizations and Improvement proposals that form our approach. These artifacts are intended to help an engineer to understand the execution of a transformation and the causes of performance problems with the help of Analyses and Visualizations based on our Monitoring in order to improve them. During the improvement the engineer will be supported by Guidelines and Improvement proposals.