Leakage Aware Scheduling on Maximum Temperature Minimization for Periodic Hard Real-Time Systems

摘要

Over the years, the chip power density has been increased exponentially due to the increasingly complicated circuit architecture as well as the continuous miniaturization of the transistor feature size. High power consumption has directly translated to high chip temperature which adversely affects the system performance/reliability and increases the cooling/packaging costs. Moreover, high chip temperature also elevates the leakage power consumption, which further augments the overall power consumption and thus the operating temperature. In this paper, we incorporate the leakage/temperature dependency as well as the nonnegligible transition overhead into analysis and present a novel real-time speed scheduling algorithm, namely M-Oscillating, that can reduce the peak temperature of a system when executing a hard real-time periodic task set. We analytically prove the correctness of the proposed algorithm based on a processor model that can effectively account for the leakage/temperature relationship. We validate the effectiveness of the proposed algorithm by comparing it with the existing work on two platforms. The first platform is a C/Matlab based chip-level thermal/power simulator, and the second platform is a more practical one built based on a desktop computer running SPEC CPU2000 benchmark programs. The experimental results obtained from both platforms demonstrate the superiority of the proposed M-Oscillating scheme over the existing approach in peak temperature reduction and feasibility improvement.