| As computer technology progresses, making smaller, lighter, and faster computing devices possible, we are experiencing a proliferation of high-performance handheld and mobile, embedded computing devices. As these devices become faster, applications more complex, and microprocessors more powerful, it is becoming difficult to provide sufficient runtime with the available stored energy in small devices. To address this, research is now starting to look into low-power and energy-conserving computer designs. Much of this work is in the domain of microprocessor and system hardware designs that reduce the energy required for computation, generally involving a trade-off between computational throughput and power consumption.; This thesis considers a software-centric approach to designing low-power embedded and mobile computing systems, identifying what must be done in the system software to create an energy-aware and -efficient system. A three-pronged approach of designing an energy-conserving OS is proposed, focusing primarily on reducing processing energy costs. (1) First, we survey various techniques of optimizing OS services to the expected characteristics of small, embedded systems, in order to reduce processing and energy overheads for the common-case scenarios. (2) Second, algorithms for real-time task scheduling are devised in order to exploit energy-conserving hardware techniques, while alleviating their performance impacts. Sprint-and-halt algorithms and real-time dynamic voltage scaling (RT-DVS) algorithms are developed to exploit software-controlled power down of hardware, and processor voltage and frequency scaling, respectively, to reduce energy consumption in real-time systems while maintaining all deadline guarantees. (3) Third, a comprehensive system of task-adaptation, called the Energy-aware Quality-of-Service (EQoS) framework is introduced. This framework provides differentiated service quality to applications tasks, weighing energy requirements against value of computation and system runtime, in order to maximize the total value of computation in a time- and energy-constrained environment.; In addition, as the energy requirements of applications is a critical input to any energy-based adaptation system, methods of measuring and modeling task energy requirements are also explored. This work investigates energy-aware operating system design by developing and implementing software mechanisms and algorithms in working systems, demonstrating the efficacy of the energy-conserving techniques developed. |
