poweraware-12 - Quantitative Comparison of Power Management...

Info iconThis preview shows pages 1–2. Sign up to view the full content.

View Full Document Right Arrow Icon
Quantitative Comparison of Power Management Algorithms Yung-Hsiang Lu, Eui-Young Chung, Tajana ˇ Simuni´c, Luca Benini, Giovanni De Micheli Computer Systems Laboratory, Stanford University, USA luyung, eychung, tajana, nanni @stanford.edu Dip. Informatica, Elettronica, Sistemistica, Universit`a di Bologna, Italy [email protected] Abstract Dynamic power management saves power by shutting down idle devices. Several management algorithms have been proposed and demonstrated effective in cer- tain applications. We quantitatively compare the power saving and performance impact of these algorithms on hard disks of a desktop and a notebook computers. This paper has three contributions. First, we build a frame- work in Windows NT to implement power managers running realistic workloads and directly interacting with users. Second, we define performance degradation that reflects user perception. Finally, we compare power sav- ing and performance of existing algorithms and analyze the difference. 1. Introduction Dynamic power management (DPM) is an effective ap- proach to reduce power consumption without signifi- cantly degrading performance [2]. DPM shuts down devices when they are not being used and wakes them up when necessary. When a device is not used, it is called idle ; otherwise, it is called busy . DPM algo- rithms observe request patterns and predict the length of idle periods. Idle periods can be defined in differ- ent ways [8]. In this paper, we consider an idle period as “time with no requests waiting for service”. The de- vice is in a working state when it can serve requests with higher power consumption, (Table 1 summaries all symbols.). The device is sleeping when it consumes less power, ( ), and cannot serve requests. Shut- ting down and waking up a device usually cause perfor- mance degradation and require extra energy. Therefore, DPM algorithms shut down a device only when an idle period is long enough to justify performance degrada- tion and state-transition energy. Design Automation and Test In Europe 2000, p 20-26 shut down wake up requests requests time before shutdown idle shutdown delay sleep time working sleeping wakeup delay working Figure 1: State Transitions This paper compares DPM algorithms for controlling the power states of hard disk drives on a desktop and a notebook computers. Hard disks are of particular interest for power management due to three reasons. First, hard disks may consume up to 20% of total en- ergy in a computer [7]. Recent studies find that hard disks will remain major power consumers in the near future [10] [12]. Second, hard disks have large perfor- mance and power overhead because of mechanical in- ertia. Spinning down or up disk plates takes several seconds, equivalent to hundreds of millions of instruc- tions in modern computers. Finally, hard disks are not always needed when computers are running if the physi- cal memory contains all the information needed; for ex- ample, caching may avoid unnecessary spin-ups [5]. Our study has three major contributions: a framework
Background image of page 1

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
Image of page 2
This is the end of the preview. Sign up to access the rest of the 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.

Page1 / 7

poweraware-12 - Quantitative Comparison of Power Management...

This preview shows document pages 1 - 2. Sign up to view the full document.

View Full Document Right Arrow Icon
Ask a homework question - tutors are online