Futuroscope-Chasseneuil

poitiers.fr This work addresses the problem of failure tolerance for real-time applications, running on a multicore hardware architec-ture. In fact, at any time during the scheduling process, a problem may occur on any of the processor cores, affect-ing the task that was running on it. We focus on systems composed of periodic independent tasks with simultaneous first release and implicit deadlines. The system is scheduled under the fair algorithm PD2. Our approach is based on limited hardware redundancy: the system will run on a processor with one more core than required. Then we prove that, if the subtask running on the faulty core is not rescheduled, the application can keep running on the remaining cores without temporal or fairness faults