An online power-efficient scheduling algorithm for multiprocessor systems

Aiming at addressing the bottleneck problem of heat dissipation in multiprocessor systems, an online power-efficient scheduling algorithm PEQUI is proposed based on the dynamic speed scaling technique. PEQUI is capable of fairly allocating processor resources by applying the dynamic partitioning strategy (EQUI). Moreover, PEQUI adjusts processor speeds in proportion to the number of active jobs by taking the non-linear relationship between processor's power consumption and its execution speed into account. Compared with traditional power-efficient algorithms, PEQUI is able to make irrevocable decisions by using only information of the current active jobs, and needs fewer decision-making parameters. Online competitive analysis and a comparison with the optimal offline algorithm show that PEQUI achieves a constant competitive ratio with respect to the total execution time and energy. Simulation results show that PEQUI achieves better performance and lower power consumption than algorithms such as the latest arrival processor sharing (LAPS) and EQUI-based strategies with constant speed. A comparison with LAPS under the same workloads shows that PEQUI effectively reduces power consumption while the execution time reduces near 7%.