Models for Deterministic Execution of Real-Time Multiprocessor Applications
Peter Poplavko1,a, Dario Socci1,b, Paraskevas Bourgos1,c, Saddek Bensalem1,d and Marius Bozga2
1University Grenoble Alpes, VERIMAG, Grenoble, F-38000, France.
2CNRS, VERIMAG, Grenoble, F-38000, France.
With the proliferation of multi-cores in embedded real-time systems, many industrial applications are being (re-)targeted to multiprocessor platforms. However, exactly reproducible data values at the outputs as function of the data and timing of the inputs is less trivial to realize in multiprocessors, while it can be imperative for various practical reasons. Also for parallel platforms it is harder to evaluate the task utilization and ensure schedulability, especially for end-to-end communication timing constraints and aperiodic events. Based upon reactive system extensions of Kahn process networks, we propose a model of computation that employs synchronous events and event priority relations to ensure deterministic execution. For this model, we propose an online scheduling policy and establish a link to a well-developed scheduling theory.We also implement this model in publicly available prototype tools and evaluate them on state-of-the art multi-core hardware, with a streaming benchmark and an avionics case study.
Full Text (PDF)