Energy-Aware Partitioning for Multiprocessor Real-Time Systems

• Authors: Hakan Aydin;Qi Yang
• Venue: IPDPS '03 Proceedings of the 17th International Symposium on Parallel and Distributed Processing
• Year: 2003

In this paper, we address the problem of partitioning periodic real-time task in a multiprocessor platform by considering both feasibility and energy-awareness perspectives: our objective is to compute the feasible partitioning that results in minimum energy consumption on m \le 2 processors. On each processor, we use variable voltage scheduling technique to adjust the voltage/speed to the load conditions, while committing to earliest deadline first scheduling policy. We show that the problem is NP-Hard in the strong sense on m \le 2 processors even when feasibility is guaranteed a priori that is, when restricted to task sets that are guaranteed to be feasible on a single processor. Then, we develop our framework where load balancing plays a major role in producing energy-efficient partitioning . We show that "heavy" tasks whose utilization values exceed a certain threshold must be assigned to separate processors in optimal solution. Then we evaluate the feasibility and energy-efficiency performances of well-known partitioning heuristics experimentally. We show that when the algorithms have the complete knowledge about the task set, Worst-Fit-Decreasing algorithm exhibits a clear superiority over other well-known techniques. However, when tasks arrive dynamically and the scheduler has to allocate tasks in a given order, experiments show that there is no technique that dominates others throughout the load spectrum. For these settings, we propose an efficient heuristic called RESERVATION, in which we combine the ideas developed in load balancing framework. Our simulation results indicate that RESERVATION offers a much more consistent performance with respect to other heuristics.