Experiences in Managing Energy with ECOSystem

Abstract

Mobile devices are becoming increasingly popular, from laptops, PDAs, and cell phones to emerging platforms such as wireless sensor networks. Available battery energy has become a critical mobilesystem resource. A mobile device's usefulness is often limited not by its hardware's raw speed but by its battery's energy.Ideally, designers should address the energy problem at all system levels: hardware, operating system, and application. At the hardware level, low-power circuit design can reduce energy consumption. In addition, hardware devices can provide device-specific power management features that you can exploit with power-management policies at the system's higher levels. At the OS level, we can observe the applications' and devices' combined resource demands. Existing OSlevel energy management tends to focus on individual devices. For example, scheduling for processors with dynamic voltage scaling (DVS) allocates CPU time to tasks and manipulates the CPU power states. [1][2][3][4] Similarly, disk and network policies concentrate only on the request patterns to their specific devices and on managing their available power states. [5][6][7][8] At the application level, you can save energy by making applications energy aware. An energy-aware application can decrease its power consumption by reducing its activity, and it can give hints to the device manager or change its device-access pattern to create energy-saving opportunities for hardware. 9,10 We designed our ECOSystem (Energy-Centric Operating System) prototype to manage energy consumption at the OS level, complementing existing power-management techniques, such as DVS and application adaptation. It's based on the ideas that energy management should be a system-wide effort, that we should explicitly recognize energy as a resource, and that we should unify energy management across the system. Even managing one hardware device might require coordination with other system components. Without unified management, application-level energy-saving efforts might not result in reduced energy consumption. ECOSystem's backgroundTwo choices initially framed our project. First, we chose to explore energy management for general-purpose applications that weren't modified to be energy aware. This seemed the most realistic scenario to ensure wide applicability of OSlevel energy management. However, it was also limiting because we'd have less information about the workload to exploit (such as deadlines, if we Energy consumption is a major systems-design challenge. ECOSystem incorporates the "currentcy model," which lets the operating system manage energy as a first-class resource. It can also express complex energy-related goals and behaviors, leading to more effective, unified management policies.

An application-aware approach to efficient power management in mobile devices

“…The reason is that the high-level applications are the ones that know best what their needs are. This information is essential to improve the energy management [4], [7], [8], [9].…”

“…Some approaches are FIS [11], ECOSystem [7], Odyssey [4], and STPM [17]. One problem is that these techniques require the modification of the Operating System.…”

“…There are several approaches which can provide adaptation as a common service or as a domain-specific service. Puppeteer [12], PAGODA [15] and PAWP [7] are domain-specific middleware whereas Spectra [13] and PARM [14] are common services middleware.…”

A Gray-Box Feedback Control Approach for System-Level Peak Power Management

“…To address power management on system-level, Lu et al [29] presented a power reduction technique at OS-level using task-based power management. Zeng et al [43] [42] proposed to build an explicitly energy-aware operating system by introducing a systemwide abstraction for the energy used. The purpose was to budget the energy available to individual processes.…”

An architecture for energy management in wireless sensor networks

“…Building systems that manage energy as a critical resource is not a new concept; research in a number of areas harnesses this idea. In fact, our architecture incorporates many of these concepts, including classifying energy as a first-class OS resource [26], prioritizing resource requests [2], accounting for fine-grained energy consumers [19], allocating resources based on dynamic constraints [7], and providing quality-ofservice (QoS) guarantees by using feedback [13]. In addition, we adopt a three component decomposition that is common for architectures managing scarce shared resources, seen in Figure 1: (1) a policy interface for user input, (2) a mechanism to monitor and control system and component resource usage, and (3) a management module for enforcing policy directives.…”