Using workflows to control the experiment execution in modeling and simulation software

To control the computation of an experiment means at least to deal with the generation of parameter combinations of interest (at best using experiment design solutions) and to execute, in case of stochastic simulations, the required replications using the hardware available. Although the process to be supported by modeling and simulation (M&S) software is often constrained to loading the model and instantiating it using a given parameter configuration and executing it by taking into account the simulation parameters, needs might arise to adapt and extend the code controlling the execution, i. e., incorporating feedback loops from validation or analysis steps into the execution scheme during runtime, having user interaction or simply adding further steps to the execution scheme such as further data processing. Additional features, like supporting diverse hardware setups, documentation or security, may imply further changes to the code. A workflow-based execution abstracts from concrete implementations and hard-coded execution patterns, as it provides a declarative description of this process, which means that anyone can set up own experimentation workflows (by using smaller predefined workflows and non-workflow-based components). Herein we present a workflow driven realization of an experimentation layer, which supports the same features as the hard-coded alternative and we discuss the pros, cons, and performance of the workflow-based approach.