Preemptive scheduling of dependent periodic tasks modeled by synchronous dataflow graphs
Résumé
Advanced features in modern cars have increased the criti-cality level of embedded applications in automotive. These applications are generally composed of several communicating functions, for which a deterministic data exchanges is crucial. In the industry, applications are designed with high level models such as Matlab/Simulink. They are implemented on an AUTOSAR platform, where they are scheduled with a fixed-priority based Operating System (OS). However, AUTOSAR OS does not directly provide support for deterministic dataflow implementation. In this paper, we present an approach to implement a deterministic dataflow of dependent periodic tasks on pre-emptive fixed-priority based uniprocessor. We consider a multi-periodic system consisting in several dependent real-time tasks modeled by a Synchronous Dataflow Graph. We use the scheduling of the graph to make the dependent tasks set independent. This permits to insure a deterministic dataflow without requiring synchronization mechanisms. In addition, it allows to use the existing scheduling policies for independent tasks. We propose several heuristics which find a scheduling solution in 76 percent of cases and provide a fast method to deal with dependencies in multi-periodic systems .