Power Aware Scheduling For Real Time Embedded Systems

Driven by the remarkable evolution of IC technology and the ever-increasing human appetite for higher computing power, the dramatically increased power/energy consumption for real-time embedded systems has presented a profound challenge to researchers and developers. Battery-operated embedded devices, which have already been ubiquitous, demand low power consumption to extend the battery life and thus the mission cycles. Even for power-rich platforms, rapidly elevated power consumption raised serious concerns regarding the reliability and packaging/cooling cost as a result of the heat dissipation. It is fair to say that energy reduction has become one of the most critical design issues in the design of next generation real-time embedded systems. In our research, we seek to address this problem at the operating system level. Specifically, we believe that real-time scheduling plays a critical role in power/energy reduction not only because most embedded systems have real-time requirements, but also because significant energy savings can be achieved by taking advantage of the knowledge in application characteristics and underlying architectures known at this level. The goal of our research is to study and develop appropriate real-time scheduling techniques that can exploit the advanced power manageable features in state-of-the-art architecture to minimize the power/energy consumption while satisfying other design requirements at the same time. The contributions of the dissertation include: (i) We developed several advanced power-aware scheduling algorithms for hard real-time systems with emphasis on reducing both dynamic and leakage power consumption; (ii) We extended the system model from simple hard real-time systems to soft real-time systems with more complicated Quality of Service constraints; (iii) We also developed efficient scheduling algorithms to minimize the system-wide energy consumption with peripheral devices taken into consideration. Experimental results have demonstrated that our techniques greatly outperform existing ones. The problems discussed in this dissertation are rather general in real-time embedded system designs, and these methodologies and techniques are important both in the theoretical and practical sense.