Towards reproducible simulation studies with JAMES II

Knowing how scientific results are obtained, meaning to know which methods, techniques, algorithms, software etc. were used, is important for those scientific results to be credible. Documentation and provenance data provide exactly this information. However, providing documentation and provenance data is not trivial. In the domain of Modeling & Simulation a multitude of different building blocks take part in conducting a simulation study. Those blocks range from software products, over mathematical models, to analysis and interaction methods. Keeping track of all this information manually is nearly impossible. Thus, a suitable computer-based support is needed. This thesis proposes a workflow-based approach for conducting simulation studies, using software aided automated documentation of workflow executions. In order for this to work a simulation study is divided into two layers, one dealing with the creation, verification and validation of a simulation model while the other deals with the execution of a simulation experiment with this model. In this thesis workflows for both layers, covering a broad range of different types of simulation studies are presented and a framework for executing those workflows as well as automatically collecting provenance data and documentation is developed and implemented. Additionally, an approach is presented and evaluated that aims at improving workflow execution in distributed environments by adaptively reacting to environmental changes optimizing execution, e.g., by speed and resource usage, utilizing machine learning techniques.