A Thermal-Aware Scheduling Algorithm for Core Migration in Multicore Processors

Multicore processors traditionally achieve increased throughput as compared to their single core counterparts by trading off area and speed performance. The reduced gate lengths that accompany process technology scaling increase the leakage power dissipation and, consequently, lead to higher power densities and thermal hotspots. Reduction of power densities and thermal hotspots is of utmost importance for proper long-term operation of the processor. In this paper, we propose a scheduling algorithm to migrate the jobs running on one processor core to other cores when the temperature in the former reaches a pre-defined threshold value. The proposed algorithm takes into account the temperature of each core identified as a prospective target for migration as well as the proximity of the prospective core to the core which is selected for job migration. The proposed algorithm is also capable of identifying the possible thermal coupling between neighboring cores. Significant temperature reduction can be achieved through the proposed scheduling and job migration by trading off some speed performance: A 6 degrees C temperature reduction was observed with a 2% delay penalty for the scheduling approach employed in this work.