Efficient Distributed Execution of Multi-component Scenario-Based Models

In scenario-based programming (SBP), the semantics, which enables direct execution of these intuitive specifications, calls, among others, for synchronizing concurrent scenarios prior to every event-selection decision. Doing so even when the running scenarios are distributed across multiple physical system components, may degrade system performance or robustness. In this paper we describe a technique for automated distribution of an otherwise-centralized specification, such that much of the synchronization requirement may be relaxed. The technique calls for replicating the entire scenario-based executable specification in each of the components, locally transforming it in a component-specific manner, and reducing the synchronization requirements to very specific and well-defined points during execution. Our evaluation of the technique shows promising results. Given that relaxed synchronization can lead to what appears as different runs in different components we discuss various criteria for what would constitute acceptable differences, or divergence, in the parallel, distributed runs of almost-identical copies of a single specification. This paper incorporates and substantially extends the material of the paper published in MODLESWARD’17 Distributing Scenario-Based Models: A Replicate-and-Project Approach by the same authors [37].