Thermal-Aware Feedback Control Scheduling for Soft Real-time Systems

Thermal management has become a prominent challenge for real-time systems as power consumption continues to increase dramatically. On the other hand, the real-time performance is seriously limited by the thermal issue because the processors have to avoid overheating all the time. As a result, a lot of precious computing resource is unnecessarily wasted. In order to address these problems, this paper presents a Thermal-Aware Feedback Control Scheduling (TAFCS) for soft real-time systems by applying feedback control theory in scheduling. TAFCS is based on a nested feedback control loop structure, where the outer control loop is for thermal control and the inner one is for miss rate control. In this way, TAFCS dynamically controls both the temperature and the real-time performance through Dynamic Voltage Scaling (DVS). The scheduling algorithm we propose can be applied in both period and aperiodic soft real-time systems. Compared with feedback miss rate control algorithm, TAFCS can more effectively improve the real-time performance, especially for the case when the temperature is much lower than the temperature reference. Compared with the FTCS algorithm, our proposed algorithm can reduce the maximum miss rate by 29.9% and the average miss rate by up to 10.4%.