Robust Partitioned Scheduling for Real-Time Multiprocessor Systems

Abstract : In this paper, we consider the problem of fixed-priority partitioned scheduling of sporadic real-time tasks for homogeneous processors. We propose a partitioning heuristic that takes into account possible Worst Case Execution Time (WCET) overruns. Our goal is to maximize the duration a task can be allowed to exceed its WCET without compromising the timeliness constraints of all the tasks. This duration is denoted in the paper the allowance of the task and is computed with a sensitivity analysis. The partitioning heuristic we propose, assigns the tasks to the processors in order (i) to maximize the allowance of the tasks and (ii) to tolerate bounded execution duration overruns. Property (ii) is important as real-time applications are often prone to be subject to OS approximations or software faults that might result in execution duration overruns. We show with performance evaluations that Allowance-Fit-Decreasing partitioning improves the temporal robustness of real-time systems w.r.t. classical {First-Fit/Best-Fit/Next-Fit}-Decreasing partitioning.