Model-as-You-Go for Choreographies: Rewinding and Repeating Scientific Choreographies

Abstract

Scientists are increasingly using the workflow technology as a means for modeling and execution of scientific experiments. Despite being a very powerful paradigm workflows still lack support for trial-and-error modeling, as well as flexibility mechanisms that enable the ad hoc repetition of experiment logic to enable, for example, the convergence of results or to handle errors. In this respect, in our work on enabling multi-scale/field (multi-*) experiments using choreographies of scientific workflows, we contribute a method comprising all necessary steps to conduct the repetition of choreography logic across all workflow instances participating in a multi-* experiment. To realize the method, we contribute i) a formal model representing choreography models and instances, including the re-execute and iterate operations for choreographies, and based on it ii) algorithms for determining the rewinding points, i.e., the activity instances where the rewinding has to stop and iii) enable the actual rewinding to a previous execution state and repetition of the choreography. We present the implementation of our approach in a message-based, service-oriented system that allows scientists to model, control, and execute scientific choreographies as well as perform the rewinding and repeating of choreography logic. We also provide an evaluation of the performance of our approach.