Reducing Total Energy for Reliability-Aware DVS Algorithms

Power aware scheduling is of increasing importance in real-time system design, especially in this global warming era. Reliability is also very critical for real-time system design. In this paper, we aim at minimizing total energy while guaranteeing reliability constraints. Total energy refers to the sum of static and dynamic energy. (Dynamic Voltage Scaling ) DVS is usually used for reducing dynamic energy consumption by reducing speed. Unfortunately, it has been shown that the transient faults of the system will be increased when the processor runs at reduced speed. To guarantee reliability be at least as high as that of without speed scaling, previous reliability aware DVS algorithms reserve recovery job for each of the scaled down tasks. However, these previous reliability aware DVS algorithms do not explore the shutdown technique to reduce static energy consumption. Static energy is consumed whenever the processor is on, and in modern processors, static energy consumption is comparable to the dynamic energy consumption and can not be ignored anymore. To lower total energy consumption, we integrate leakage control method and shared-recovery technique with reliability aware DVS algorithms. Experimental results show that our methods are effective.