| The emergence of mobile systems has been a critical driving factor for low power design. This dissertation addresses operating system (OS) design of mobile systems targeting energy efficiency. It first highlights the energy-efficiency bottleneck in the form of leakage power that is emerging as the technology feature size shrinks. To tackle this bottleneck, it presents an analytical energy model which considers dynamic and leakage power simultaneously. The model combines dynamic voltage scaling (DVS) and adaptive body biasing (ABB) to consider the trade-off between supply voltage and threshold voltage for a given performance level. Based on this model, it presents a leakage-aware real-time scheduling algorithm to maximize energy savings without violating real-time constraints. It then exploits the opportunities for energy savings by considering user experience for interactive applications. It proposes the use of user-perceived latency, i.e., the delay between user input and computer response, for directly driving voltage scaling. To improve energy efficiency without harming user experience during voltage scaling, it analyzes multiple contributors to user-perceived latency and proposes a holistic approach to user focus-aware resource management that addresses process scheduling, memory management, disk I/O scheduling, and voltage scaling in the OS, and X client scheduling in the X Window system. Finally, it presents a computing model for wireless device drivers employed in wireless personal area networks (WPAN), which explores the potential for energy savings by taking user experience into account. It offers an extensive comparative study of two widely-used WPAN technologies, Bluetooth and ZigBee, in terms of performance and energy efficiency. |
