ポテンシャルゲームによるMP-SoCの temperature-aware な周波数最適化

抄録

Recently, an architecture called multiprocessor system-on-a-chip (MP-SoC) is paid much attention to. MP-SoC is composed of several processing elements (PEs) interconnected with a network-on-a-chip. It is important for MP-SoC not only to reduce the computational time of the executed task but also to avoid the rise of the temperature. In this paper, we formulate an optimization problem of clock frequencies of PEs where an objective function evaluates both the computational time and the temperature of MP-SoC. We transform the optimization problem into a potential game and propose a distributed algorithm named a round-robin update algorithm for optimizing the frequencies of PEs. By using simulations, we also compare the proposed algorithm with an existing algorithm and show advantages of the proposed algorithm.