Temperature-Aware Distributed Run-Time Optimization on MP-SoC Using Game Theory

With forecasted hundreds of processing elements (PE), future embedded systems will be able to handle multiple applications with very diverse running constraints. In order to avoid hot-spots and control the temperature of the tiles, dynamic voltage-frequency scaling (DVFS) can be applied at PE level. At system level, it implies to dynamically manage the different voltage-frequency couples of each PE in order to obtain a global optimization. In this article we present an approach based on game theory, which adjusts at run-time the frequency of each PE. It aims at reducing the tile temperature while maintaining the synchronization between the tasks of the application graph. A fully distributed scheme is assumed in order to build a scalable mechanism. Results show that the proposed run-time algorithm find solutions in few calculation cycles achieving temperature reductions of about 23%.