This preview shows pages 1–2. Sign up to view the full content.
This preview has intentionally blurred sections. Sign up to view the full version.View Full Document
Unformatted text preview: Operating-System Directed Power Reduction Yung-Hsiang Lu CSL, Stanford University, USA [email protected] Luca Benini DEIS, Universit`a di Bologna, Italy [email protected] Giovanni De Micheli CSL, Stanford University, USA [email protected] ABSTRACT This paper presents a new approach for power reduction by taking a global, software-centric view. It analyzes the sources of power consumption: tasks that require services from hardware components. When a component is not used by any task, it can enter a sleeping state to save power. Operating systems have detailed information about tasks; therefore, OS is the best place for power reduction. Our technique is effective in identifying hardware idleness and shutting down unused components. We implement this tech- nique in Linux and show that it can save more than 50% power compared to traditional hardware-centric shutdown techniques. 1. INTRODUCTION Low-power design is increasingly important because of the popularity of portable devices and the concern of environ- mental impact of electronic systems . Due to rapid advance in hardware, modern systems are supporting wide ranges of applications. Specifically, computers and some PDA’s deploy operating systems, such as Windows, Palm OS and Linux to manage resources. This paper focuses on such systems. In many systems, some hardware components (also called devices ) are not always used; workload variations make adap- tive power reduction possible. These techniques reduce power consumption according to the requests generated from run- ning tasks. If no task requires a particular device, the device is idle and can be shut down to enter a low-power low-performance sleeping state (also called stand-by state ). When a running task requires this device, it is woken up and enters a high-power high-performance working state . Dy- namically determining power states according to workloads is called dynamic power management (DPM) . Power managers (PM) determine power state transitions accord- ing to their shutdown rules (also called policies ). Power management can be generalized for more than two power states ; each state has different performance and power consumption. Therefore, power management includes dynamic voltage setting  and variable clock speeds . Setting voltages or clock speeds is equivalent to choosing power states. We use “power management” for all tech- niques that dynamically reduce power based on workloads. For simplicity of the following presentation, we consider only two power states: working and sleeping. Power management in commercial products is mainly based on timeout— shutting down a device when it has been idle long enough. Most research activities take a hardware-centric view; they observe past requests at the target device to pre- dict future idleness   . Some schemes use stochas- tic models due to the lack of information to distinguish re- questers   . None of these approaches makes distinction on the request source; they implicitly assume...
View Full Document
This note was uploaded on 09/01/2009 for the course CSE CS-699 taught by Professor Prf.p.bhaduri during the Spring '09 term at Indian Institute of Technology, Guwahati.
- Spring '09