1 On the Energy Efficiency of Device Discovery in Mobile Opportunistic Networks: A Systematic Approach Bo Han∗, Jian Li† and Aravind Srinivasan‡ ∗AT&T Labs – Research, 1 AT&T Way, Bedminster, NJ, 07921 †Institute for Interdisciplinary Information Sciences, Tsinghua University, Beijing 100084, China ‡Department of Computer Science and the Institute for Advanced Computer Studies, University of Maryland, College Park, MD, 20742 Abstract—In this paper, we propose an energy efficient poses. For example, LoKast1 is an iPhone application that device discovery protocol, eDiscovery, as the first step to provides mobile social networking services by discover- bootstrapping opportunistic communications for smartphones, ing and sharing media content among users in proximity. the most popular mobile devices. We chose Bluetooth over 2 WiFi as the underlying wireless technology of device discovery, Nintendo 3DS’s StreetPass enables players to exchange based on our measurement study of their energy consumption game data with other users they pass on the street, through on smartphones. eDiscovery adaptively changes the dura- the direct device-to-device communication between 3DS tion and interval of Bluetooth inquiry in dynamic environ- systems. Other similar applications include Sony PS Vita’s ments, by leveraging history information of discovered peers. Near and Apple’s iGroups. We implement a prototype of eDiscovery on Nokia N900 smartphones and evaluate its performance in three different Device discovery is essentially the first step of oppor- environments. To the best of our knowledge, we are the first to tunistic communications. However, there are very few prac- conduct extensive performance evaluation of Bluetooth device tical protocols proposed for it and most of the existing work discovery in the wild. Our experimental results demonstrate mainly utilizes (trace-driven) simulation to evaluate the that compared with a scheme with constant inquiry duration performance of various device discovery protocols [9], [31]. and interval, eDiscovery can save around 44% energy at the expense of discovering only about 21% less peers. Moreover, although there are several real-world mobility 3 The results also show that eDiscovery performs better traces in the CRAWDAD repository which were collected than other existing schemes, by discovering more peers and using Bluetooth device discovery, most of them used very consuming less energy. We also verify the experimental results simple discovery protocols with fixed inquiry duration through extensive simulation studies in the ns-2 simulator. and interval. A recently proposed opportunistic Twitter Index Terms—Device discovery, opportunistic communica- application [26] also uses a 2-minute inquiry interval for tions, energy efficiency, smartphones, Bluetooth. Bluetooth device discovery. It is known that these kinds of discovery protocols are not energy efficient [31] and thus may not be desirable for power-constrained mobile devices, I. INTRODUCTION such as smartphones. In this paper, we bridge this gap by developing an energy-aware device discovery protocol Mobility itself is a significant problem in mobile net- for smartphone-based opportunistic communications and working. On the one hand, protocols designed for mobile evaluating its performance in practice. networks should solve the challenges caused by the mo- There are two major challenges in designing, imple- bility of wireless devices. For example, routing protocols, menting and evaluating energy efficient device discovery such as DSR (Dynamic Source Routing) [16], are required protocols for smartphones. First, the selection of underlying to handle frequent routing changes and reduce the cor- communication technology is complicated by the multiple responding communication overhead. On the other hand, wireless interfaces on smartphones, such as Bluetooth and mobility can increase the capacity of wireless networks WiFi (a.k.a., IEEE 802.11).4 Although Bluetooth is a low- through opportunistic communications [14], where mobile power radio, its device discovery duration is much longer devices moving into wireless range of each other can than WiFi (∼10s for Bluetooth vs. ∼1s for WiFi active exchange information opportunistically during their periods scanning), which may cause more energy consumption on of contact [7], [21]. smartphones. Similarly, WiFi is known to be power-hungry Opportunistic communications have been widely ex- for mobile devices [24], [28]. Thus, it is not clear which of plored in delay-tolerant networks [34], mobile social ap- 1 plications [21] and mobile advertising [1], to facilitate http://www.lokast.com/ 2http://www.nintendo.com/3ds/features/ message forwarding, media sharing and location-based ser- 3http://crawdad.cs.dartmouth.edu/ vices. Meanwhile, there are more and more applications 4We prefer Bluetooth and WiFi to 3G, as they are local leveraging opportunistic communications for various pur- communication technologies with almost no monetary cost. 2 them is more suitable for device discovery on smartphones. A. Wireless Device Discovery in General Second, given the dynamic nature of human mobility, we Device discovery has been widely studied in various need to adaptively tune the parameters of device discovery, wireless networks, such as ad-hoc networks [20], [30], such as inquiry duration and interval, to reduce smartphone mobile sensor networks [10], [17] and delay-tolerant net- energy consumption. Schemes with constant inquiry inter- works [31]. vals have been proven to be optimal in terms of minimizing discovery-missing probability [31]. However, their energy Neighbor/device discovery is one of the first steps to ini- consumption is usually higher than the adaptive ones, tialize large wireless networks. McGlynn and Borbash [20] which may miss more devices during discovery procedures. examine the problem of neighbor discovery during the Therefore, there is a tradeoff between energy consumption deployment of static ad-hoc networks, where the discovery and discovery-missing probability. may last only a few minutes. Inspired by the birthday paradox, a pair of nodes perform neighbor discovery by We make the following contributions in this paper. transmitting and listening on k independently and randomly • We present a systematic measurement study of the chosen slots among n slots (the ratio k/n is relatively energy consumption of Bluetooth and WiFi device small). Vasudevan et al. [30] show that an existing ALOHA- discovery on smartphones, by measuring both the elec- like neighbor discovery algorithm reduces to the classical trical power and the discovery duration (Section IV). Coupon Collector’s Problem when nodes are not capable Based on our measurement results, we chose Bluetooth of collision detection. They also propose an improved as the underlying wireless technology. We emphasize algorithm based on receiver status feedback when nodes that although previous works have studied the power have a collision detection mechanism. Differently from the of Bluetooth/WiFi devices [9], [11], [24], they either above works that are based on abstract communication focus on only Bluetooth [9] or ignore the duration of models, our focus is practical Bluetooth device discovery device discovery [11], [24], without which it is hard for smartphone-based opportunistic communications. to evaluate the energy consumption of these devices. Dutta and Culler [10] propose an asynchronous neighbor • We design an energy-aware device discovery protocol, discovery protocol, called Disco, for mobile sensing appli- named eDiscovery, as the first and very important cations. Disco can address the challenge of operating the step to bootstrapping smartphone-based opportunistic radios at a low duty cycle and ensuring fast and reliable communications (Section VI). By trading energy con- discovery in bounded time through the adaptation of the sumption for a limited discovery loss, we demonstrate Chinese Remainder Theorem. U-Connect [17] is another that eDiscovery is highly effective in saving energy asynchronous neighbor discovery protocol for mobile sen- on smartphones. eDiscovery dynamically tunes the sor networks that selects carefully the time slots to perform discovery duration and interval according to history discovery and that has been proven theoretically better than information of the number of discovered peers. It Disco. Recently, Bakht et al. [2] propose Searchlight, a also introduces randomization into device discovery, protocol that combines both deterministic and probabilistic in order to explore the search space further. approaches to further reduce the discovery latency for • Our major contribution is an extensive performance mobile social applications. Disco, U-Connect and Search- evaluation of eDiscovery and other existing device light mainly aim to achieve a tradeoff between discovery discovery protocols in different realistic environments, latency and energy consumption. For example, U-Connect through a prototype implementation on Nokia N900 uses the power-latency product metric for performance smartphones (Section VII). We conduct experiments in evaluation. Differently from them, we are interested in a university campus, a metro station and a shopping the tradeoff between energy consumption and discovery- center. Our experimental results verify the effective- missing probability. eDiscovery ness of in practice. Compared with Cohen and Kapchits [6] investigate a slightly different the STAR protocol proposed by Wang et al. [31], neighbor discovery