Non-preemptive multiprocessor scheduling for periodic real-time tasks

Abstract

Scheduling of a set of tasks in multiprocessor environment is a computationally intensive job. There are primarily two broad approaches for scheduling of tasks on multiprocessors. In one of the approaches tasks are allocated to processor in the beginning (partition based strategy) and the other approaches maintain a global scheduler. There exist a large volume of work in multiprocessor scheduling having different optimization objectives such as schedule length, response time, processor utilization, etc. However, most of these works focus only in preemptive scheduling. Very less attention has been given for scheduling of non-preemptive tasks. In this work, we present a methodology for scheduling of a set of non-preemptive real-time tasks using minimum number of processors. We consider that the tasks are allocated to the processors using bin-packing strategies such as firstfit or best-fit and non-preemptive Earliest Deadline First (npEDF) scheduling method is applied to each processor. We did extensive experiments by combining different partitioning strategies with the ordering of tasks (period, utilization, etc). We found that First-Fit Increasing Period, Best-Fit Increasing Period, First-Fit Decreasing Utilization and Best-Fit Decreasing Utilization give reasonably good results. The success ratio of decreasing utilization heuristics are 10%–30% more than increasing period heuristics. We observed that First-Fit Decreasing Utilization and BestFit Decreasing Utilization takes more time than First-Fit Increasing Period and Best-Fit Increasing Period. We also compared our results with the existing approach.