Thermal coupling aware task migration using neighboring core search for many-core systems

Task migration is one of the promising dynamic thermal management techniques for multi-core and many-core architectures. In this paper, we observe the conventional hot-cold swap policy becomes not effective as the core size shrinks, and the thermal coupling effect between nearby cores plays a vital role to avoid the ping-pong phenomenon (tasks being migrated back and forth). We propose a neighboring core search based migration to find the core-pair candidate subject to the lowest thermal coupling effect. Experimental results show that our approach has significantly less overheating time, unbalanced time, and migration overhead compared with the state-of-the-art neighboring task migration technique at the expense of negligible processing throughput degradation.