Collaborative power management: Piercing abstraction barriers for fast and energy-efficient pervasive data access

A significant problem with current I/O power management is that it is device-centric: each device implements its own algorithm without considering the context in which it is operating. This dissertation puts forth the claim that devices should adapt their power management to their environment for fast and energy-efficient data access. Such adaptation can be achieved through a collaboration between users, applications and devices in a mobile computing environment. This approach is called collaborative power management.; This dissertation begins by improving the power management for individual I/O devices. It describes the design and implementation of self-tuning power management (STPM) for network and storage devices. STPM provides simple interfaces that allow applications and users to share context with device drivers and power management algorithms that effectively utilize this external context.; The dissertation next shows how applications can improve power management by coordinating their data, access across multiple I/O devices. It provides the design and implementation of interfaces that allow devices to share information about their current state and characteristics with adaptive applications. Adaptive applications use this context to dynamically select an I/O device to access data for better performance and energy-savings. It also extends the STPM interface to allow adaptive applications to provide passive hints about their data access intent even when they are employing another device for data access. This support allows devices to adapt better to application needs.; The final part of this dissertation describes the design and implementation of PAN-on-Demand; a system that allows co-located computers to collaborate with each other for better data access. PAN-on-Demand provides an energy-efficient infrastructure that allows co-located computers to self-organize in a wireless personal-area network (PAN) and share data using multiple radios, such as Bluetooth and Wi-Fi. When applications are actively using the network, PAN-on-Demand offers high-bandwidth, low-latency communication; when demand is light, PAN-on-Demand adapts the network structure to minimize energy usage.