Energy-aware dynamic slack allocation for real-time multitasking systems

Dynamic voltage scaling (DVS) has been a very effective technique for processor energy reduction. It adjusts processor voltage and frequency level during runtime. In this article, we propose a general and flexible processor voltage scaling algorithm for real-time multitasking systems. Our approach focuses on exploiting dynamic slack that is created when a task finishes earlier than its estimated worst-case execution time (WCET). Our algorithm is efficient enough to execute at runtime and can be configured flexibly to make tradeoffs between running time and energy savings. By rescheduling tasks effectively, we can achieve almost as much energy savings as if there is no arrival time constraints. Furthermore, our approach can effectively incorporate both leakage power consumption as well as variable scaling overhead. Also, it is relatively independent of task characteristics and scheduling policy. Experimental results show that our technique can achieve significant energy savings at runtime over statically generated schedules and up to 12% more savings compared to the state-of-art techniques.