866 IEEE TRANSACTIONS ON MULTIMEDIA, VOL. 8, NO. 4, AUGUST 2006 Correspondence A Client-Side Statistical Prediction Scheme for Energy Aware Multimedia Data Streaming Yong Wei, Suchendra M. Bhandarkar, and Surendar Chandra Abstract—The recent proliferation of streaming multimedia on a variety of mobile devices has severely tested their battery lifetime. The long run- ning nature of typical streaming applications results in significant energy consumption by the wireless network interface card (WNIC) in these mo- bile devices. In this paper we explore linear prediction-based client-side strategies that reduce the WNIC energy consumption to receive multimedia streams by judiciously transitioning the WNIC to a lower power consuming sleep state during the no-data intervals in the multimedia stream, without explicit support from the multimedia servers themselves. Experimental re- sults on popular streaming formats such as Microsoft Media, Real and Apple QuickTime show that a linear prediction-based strategy performs better than history-based strategies that use simple temporal averaging. Index Terms—Energy-aware computing, linear prediction, mobile com- puting, multimedia streaming. I. INTRODUCTION The recent proliferation of multimedia capable mobile computing devices and networking technologies have created enormous opportu- nities for mobile device users to communicate with one another using multimedia streams. A necessary criterion for the mass acceptance of mobile devices is acceptable battery life of these devices. There has been dramatic improvement in energy-aware design of systems, both, in terms of hardware and software. Unfortunately, advances in hard- ware and software are not matched by a corresponding increase in battery life. Thus, the usefulness of these mobile devices in watching and/or hearing streaming multimedia is restricted by battery capacity. Future trends in battery technology do not promise dramatic improve- ments in battery capacity that will make this issue disappear. Conse- quently, hardware or software solutions need to be developed at the system or application level to prolong battery life. Previous work on power management for mobile devices includes spin-down policies for disks [5]–[8], scheduling policies for reducing CPU energy consumption [9], [10] and managing wireless communi- cations [11]–[14]. An IEEE 802.11b Wi-Fi connection is a popular way for mobile consumers to access the Internet wirelessly. The en- ergy consumption of the wireless network interface can be significant, especially for smaller devices. Since media streaming applications are typically long running, the power consumption of these applications needs to be taken care of. Early work by Stemm et al. [2] reports that the network interface draws a significant amount of power. Although Manuscript received May 11, 2004; revised November 3, 2005. This work was supported in part by a research grant from the State of Georgia Yamacraw Program in Embedded Software Systems to S. M. Bhandarkar and S. Chandra. The associate editor coordinating the review of this manuscript and approving it for publication was Dr. Chang Wen Chen. Y. Wei and S. M. Bhandarkar are with the Department of Computer Sci- ence, University of Georgia, Athens, GA 30602 USA (e-mail: yong@cs.uga. edu; suchi@cs.uga.edu). S. Chandra is with the Department of Computer Science and Engineering, University of Notre Dame, Notre Dame, IN 46556 USA (e-mail: surendar@ cse.nd.edu). Digital Object Identifier 10.1109/TMM.2006.876232 Fig. 1. Energy consumption rates of two WNIC’s in various states. dependent on the specific machine and wireless device, the energy con- sumption of wireless communication devices can represent over 50% of total system energy consumption for current handheld computing devices and up to 10% for high-end laptops [2]. Feeney et al. [18] also report the energy consumption measurements of an IEEE 802.11b WNIC in an ad-hoc networking environment and show that the energy consumption of the IEEE 802.11b WNIC has a complex range of be- havior. Hence, it is important to look at techniques to reduce the energy consumed by the network interface used to download the multimedia stream. The energy consumption rates of a wireless network interface card (WNIC) in the sleep state and in the receive, transmit or idle states are substantially different. Fig. 1 shows the power consumption rates of two popular WNIC’s in the various aforementioned states [3]. The WNIC’s energy consumption rate when receiving, transmitting data or when idling is substantially higher than when sleeping. Note that the WNIC cannot transmit, receive, or buffer data in the sleep state. Lorch et al. [15] present a survey of software techniques for energy management. Havinga et al. [16] present an overview of energy man- agement techniques for multimedia streams. Aggarwal et al. [17] de- scribe techniques for processing video data for transmission under low power situations. A popular strategy to reduce the energy consumption of wireless network devices is by switching them to the lower power sleep state. Systems employing a strategy which enables switching of the WNIC to a low power consumption sleep state can achieve energy savings whenever possible without modifying the underlying appli- cation and without user-visible latency. Frequent switching to a low power consumption state also promises the added benefit of allowing the batteries to recover, thus exploiting the battery recovery effect [4]. Media transcoding is a popular strategy used to reduce the stream fidelity. This strategy reduces the stream size, and hence reduces the amount of network traffic. Reducing the network traffic has the poten- tial of reducing the total energy consumed. However, if care is not taken to return the WNIC to the low power-consuming state for as often and as long as possible, reducing the amount of transmitted data will have a negligible effect on the overall client energy consumption. The basic principle underlying the proposed energy-saving approach is to predict the time durations during which to suspend communi- cation by switching the WNIC to a sleep state. Our analysis of typ- ical streams shows that the WNIC spends most of the time waiting for stream packets in a higher energy consuming idle state. Even for a high bandwidth 2000 Kbps stream, the WNIC spends over 56% of the time in the idle state; illustrating the potential for significant energy savings. Our policies operate on the multimedia client without explicit coordi- nation or help from the multimedia server. Multimedia/video data is 1520-9210/$20.00 © 2006 IEEE Authorized licensed use limited to: University of Georgia. Downloaded on November 10, 2008 at 13:21 from IEEE Xplore. Restrictions apply.