Scheduling periodic real-time hardware tasks on dynamic partial reconfigurable devices subject to fault tolerance

Reconfigurable devices employed in safety-critical real-time applications ought to tolerate faults in a timely manner. Fault-tolerant schedulers can be used to guarantee both real-time and fault-tolerant behaviors of such systems. In this paper the effects of three real-time scheduling algorithms Earliest Deadline First (EDF), Least Laxity First (LLF), and Rate Monotonic (RM) enhanced by two fault tolerance techniques Primary/Backup and Checkpointing on the correct behavior of a reconfigurable system in the presence of faults and errors are investigated in details. The obtained results confirm that RM comparatively outperforms both EDF and LLF in error recovery and incurs less allocation and preemption count as well. The results also indicate that it would be more advantageous to use Primary/Backup technique when the number of tasks, tasks size, or the fault occurrence probability is low, otherwise when the cost of saving tasks state is low, Checkpointing technique would be a good choice.