Optimising Highly-Parallel Simulation-Based Verification of Cyber-Physical Systems

Cyber-Physical Systems (CPSs), comprising both software and physical components, arise in many industry-relevant domains are often mission- or safety-critical. System-Level Verification (SLV) of CPSs aims at certifying that given (e.g., safety liveness) specifications met, estimating the value some KPIs, when system runs its operational environment, i.e., presence inputs (from users other systems) and/or additional, uncontrolled disturbances. To enable SLV complex systems from early design phases, currently most adopted approach envisions simulation a model under (time bounded) scenarios interest. Simulation-based can be computationally prohibitive (years sequential simulation), since is intensive set interest huge. We present technique that, collection (extracted mass-storage databases symbolic structures, e.g., constraint-based scenario generators), computes parallel shortest campaigns, which drive possibly large number simulators running HPC infrastructure through all (and only) those user-defined (possibly random) order, by wisely avoiding multiple simulations repeated trajectories, thus minimising overall completion time, compatibly with available simulator memory capacity. Our experiments on Modelica/FMU Simulink case study models up to ~200 million show our optimisation yields speedups as high 8x. This, together enabled massive parallelisation, makes practically viable (a few weeks infrastructure) verification tasks (both statistical exhaustive, respect scenarios) would otherwise take inconceivably long time.