Thermal-Aware Scheduling for Peak Temperature Reduction with Stochastic Workloads

In this paper we proposed a thermal-aware scheduling algorithm for peak temperature reduction with stochastic workloads. The proposed approach consists of two steps. The first step constructs a job sequence based on the mean of the execution time of each job; and the constructed job sequence interleaves the execution of hot jobs and cool jobs. The hot job pushes up the chip temperature while the cool one cools down the chip. Accordingly the chip reduces the peak temperature. The second step exploits the job slack to cancel out the adverse thermal impact due to the uncertainty of the job execution time. The slacks are allocated among the hot jobs. Hot jobs whose execution time deviates more above the average execution time will get more slack, vice versa. Experimental results show that the proposed approach effectively avoids the thermal emergency for the applications with stochastic workloads by achieving the peak temperature reduction of 4.7C on average.