Performance comparison of power-saving strategies for mobile Web access

Abstract

One of the critical issues in mobile Web access is the usage of limited energy resources of mobile computers. Unfortunately, the legacy TCP/IP architecture is very inefficient. This work proposes and analyzes power-saving strategies for mobile Web access. Specifically, in this paper we develop an energy-consumption model for Web transactions and, based on it, we propose and compare four different energy saving strategies: ideal, Indirect-TCP (I-TCP), local and global. The ideal strategy is unfeasible but it is used as a reference bound as it guarantees the lowest energy consumption. The other strategies have been implemented and compared in a real test-bed. The performance comparison is carried out by measuring two main performance figures: the energy spent for downloading a Web page, and the associated transfer-time. Experimental results show that relevant energy saving is achievable and that, among the feasible strategies, the global one gives the best performance: with this strategy we can save (on average) up to 88% of the energy. Furthermore, our results indicate that this power saving is obtained without a significant increase in the transfer-time perceived by the users (on average, 0.2 s). Finally, by comparing the feasible strategies, we observe that the global one is much closer to the ideal case than the other strategies. In detail, the global strategy is about twice more efficient than the local one, and eight times more efficient than the I-TCP strategy.

Introduction

In the field of mobile computing, the mobile Internet is one of the most interesting areas. Users are no longer forced to access information at their desktop, since data are available where they are, at any time and any place. However, several problems must be solved when integrating a mobile device in the Internet. As is well known, one of the most critical problems is the energy consumption [1], [2], [3]. In this work we consider how to introduce power saving in mobile Web access.

Efficient energy management has been approached at different levels of a mobile system architecture: physical transmission [4], [5], [6], MAC protocols [7], [8], [9], [10], [11], disk and CPU management [12], [13], [14], [15], [16], and applications [2], [17], [18], [19], [20], [21], [22]. Experimental results show that a relevant part of the energy available on a mobile computer is drained by the wireless interface. More precisely, the networking impact on energy consumption varies from about 10% in laptops [23] up to 50% in small-size hand-held devices, such as PDAs [24]. Therefore it is vital to design energy-efficient networking subsystems.

The key point in energy-aware networking is the consumption model of a wireless interface. Specifically, the wireless interface consumes nearly the same amount of energy in the receive, transmit and idle states (see, for example, the 802.11 “Wi-Fi” environment [25]). Therefore, the energy consumption is approximately proportional to the time during which the wireless interface remains switched on. The maximum power saving can therefore be achieved by transmitting data as quick as possible and, immediately after, turning the wireless interface into a power-saving mode. Many researchers have proposed power-saving policies based on this idea [23], [24], [26], [27], [28]. The innovative contribution of the approach presented in this paper is the exploitation of the application semantic to determine the best time instants for switching the wireless interface on and off.

In this paper we refine, and extensively evaluate, the power-saving architecture defined in [29]. Specifically, we compare the performance of four different power-saving strategies aimed at reducing the energy consumed during a Web-page download. The first strategy is a pure Indirect-TCP (I-TCP) architecture [30], [31]. With respect to the legacy TCP architecture, this solution improves the throughput achieved by the mobile host, thus reducing the transfer-time. Hence, it indirectly contributes to power saving even though no energy-management mechanism is explicitly introduced in the system. Explicit energy management is included in the other policies we consider, all obtained by enhancing the I-TCP architecture. The local strategy switches the wireless interface off when the user is reading the Web page, i.e., it exploits information that are locally available at the client browser. The third approach, referred to as the global strategy, in addition to local information, exploits statistical information about Web traffic. Finally, an ideal (unfeasible) strategy that guarantees the minimum power consumption is also considered. Throughout this paper, the ideal and I-TCP strategies provide the lower and upper bound for energy consumption, respectively.

We implemented the feasible power-saving strategies and tested them extensively in a real Internet scenario. Our performance study is based on two main performance figures: I_ps and I_pd. I_ps is used as a power-saving index. It measures the energy consumption of a specific strategy expressed as a percentage of the energy consumption related to I-TCP strategy. I_pd measures the impact of the power-saving strategy on the user response time (URT), i.e., the time interval elapsed from a user request for a Web page to its rendering on the mobile device.

The experimental results show that the global strategy exhibits the best achievable performance. It saves, on average, 88% of the energy consumed by the I-TCP approach and has a negligible impact on the URT (the URT increase is of 0.2 s on average, and is below 1.8 s with probability 0.9).

The paper is organized as follows. Section 2 models the statistical properties of a Web transaction and its energy consumption. Section 3 defines the power-saving strategies and their energy consumption. Section 4 presents the experimental test-bed, and introduces the performance indexes used in our measurement study. 5 Tuning of the experiments, 6 Performance evaluation analyze the performances of the different power-saving strategies. Section 7 concludes the paper.