Game-Theoretic Approach for Temperature-Aware Frequency Assignment with Task Synchronization on MP-SoC

Multiprocessor system-on-chip composed of several processing elements will be integrated in embedded systems handling multiple applications with diverse constraints. In order to improve the system performance and temperature profile, dynamic voltage frequency scaling can be applied at processor level. In this article we study an existing approach based on Game Theory, which adjusts at run-time the frequency of each processor. It aims at reducing the tile temperature while maintaining the synchronization between tasks of the application graph. We present a technique to improve the convergence of the algorithm. Our results show that 100% of scenarios converge to a solution. The optimization quality is studied in a test-case where the temperature profile and task synchronization results are compared to an off line optimization method showing similar results.