Implementation and performance evaluation of a self-timed multiprocessor scheduler on a shared-memor

Abstract

A new multiprocessor scheduling heuristic, Ranked Weight heuristic, based on the Critical Path approach is presented. The proposed heuristic is embedded in a multiprocessor scheduling algorithm that divides the tasks in a task system into a number of independent layers and schedules the resulting task sets using an urgency xule based on the proposed heuristic. Performance of the proposed scheduling algorithm is measured through extensive performance modeling studies. The behavior of the algorithm is described in terms of the number of processors required for execution of a task system and the actual number of allocated processors. The multiprocessor scheduler described in this paper is implemented on a Sequent Symmetry machine using a self-timed scheduling approach. Due to the timeshared multiuser nature of processing on Sequent, a benchmarking program is used to collect the run-time overhead associated with self-timed scheduling of task systems. The run-time behavior of task systems is analyzed and described in terms of each task system's scheduled critical pafh.