The 14th International Workshop on Mobile Computing Systems and Applications (ACM HotMobile 2013)

Sarah Clincha Marcelo Martinsb s.clinch@comp.lancs.ac.uk martins@cs.brown.edu a School of Computing and Communications, Lancaster University, Lancaster, Lancashire, UK b Department of Computer Science, Brown University, Providence, RI, USA

I.

Introduction

III.

Keynote Wearable Computing: Through the Looking Glass

The Fourteenth Workshop on Mobile Computing Systems and Applications (ACM HotMobile 2013) was held over 26th and 27th February at Jekyll Island, GA, USA. The workshop brought together approximately 90 researchers from academic and industrial institutions. A total of seventeen papers were presented during six sessions (approximately 31% of the submissions); the workshop also featured a keynote from Thad Starner (Georgia Institute of Technology), a panel on mobile systems and the developing world and a poster and demo session (thirteen posters, six demos). Each of the paper and panel sessions were followed by a discussion period. This report summarises the presentations and discussion that took place.

II.

Reoccurring Themes

Throughout the workshop the issue of user experience frequently generated discussion. In the keynote, Starner identied a need to improve interaction in order to reduce the cognitive and attention demands of mobile computing, whilst also highlighting that use of technology rarely resembled initial predictions. Concerns about cognitive load were raised again within the Vehicular Networking session. Within other sessions, balancing customisation effort against optimisation gains was a key concern, as was the provision of clear visualisations to support decision making. Another frequent discussion topic was the empowerment of new applications. Many of the systems and solutions discussed in the panel on Mobile Computing and the Developing World (VIII) used technological constraints as a motivator for the development of creative and affordable solutions to bridge the IT gap in small communities. The session on Sensors and Data (VII) also showed a variety of groundwork to facilitate the development of mobile applications, be it from the perspective of collaborative sensing, data collection, or privacy enhancement.

Mobile computing systems could easily be accused of consuming too much attention, diverting concentration from the physical environment. Thad Starners opening keynote highlighted the contrast between mobile computing and true on-the-go computing devices you can really use in motion. Starner suggested the use of micro-interactions to reduce the time between intention and action, allowing truly mobile computing. Throughout the keynote, Starner used examples from his labs to demonstrate the potential of micro-interactions. One early (pre-smartphone) example allowed Starner to conduct Web searches via his wristwatch, typically demonstrated as a parlour trick in which he offered to answer any question posed. Recently, Starner has been involved in Googles Project Glass. Playing a recent promotional video 1 , Starner showed Glasss potential to support micro-interactions for video and still photography, and voice chat, amongst others. As another example, Starner demonstrated the Mobile Music Touch (MMT), a wireless glove with vibration motors in the ngers. Wearing the glove allows an individual to learn piano ngering whilst attending to other tasks. Trials of MMT show effective learning whilst completing exams, presenting a conference paper and forecasting the weather 2 . In the healthcare domain, MMT has been shown to be an effective tool in tetraplegic rehabilitation; patients showed increased mobility and sensation, whilst the learning of a life- long skill provided incentive to continue. MMT dominated the audience questions. Degradation was one concern Starner reported that six months after their rehabilitation study, one participant played a piece for CNN just fteen minutes of practice were sufcient to refresh learning. Exploration of
http://www.google.com/glass/start/ how-it-feels/ 2 http://www.youtube.com/watch?v=uEdr6iY6F-w
1

Mobile Computing and Communications Review, Volume 17, Number 2, April 2013

the limits and alternative applications is also an interesting area. Jason Hong (Carnegie Mellon University) asked about insights gained from using the various wearable technologies. Starner responded by noting that almost everyone does not use a technology the way they initially think they will and that productivity tools often emerge perhaps wearable computing is less about the killer app and more about the killer lifestyle.

V.

Power Management

IV.

Cellular Billing

The rst session was chaired by Suman Banerjee (University of Wisconsin-Madison) and began with Splitting the Bill for Mobile Data with SIMlets [13] presented by Alec Wolman (Microsoft Research). Motivated by the limited mobile broadband infrastructure, the paper suggests the use of split billing. Split billing allows content providers to pay costs generated by mobile users using their services. SIMlets support split billing by providing a trustworthy abstraction whose policy indicates which trafc should be billed to which entity. Content providers issue a SIMlet to a mobile device in order to specify which data they are prepared to pay for. Audience members suggested that mobile operators might also be incentivised towards split billing, but that the key challenge is the development of the required trust relationships. Concerns were also raised about the scalability of the hardware-specic aspects of their approach given the rapid pace of change in current smartphones. Younghwan Go (KAIST) then presented Towards Accurate Accounting of Cellular Data for TCP Retransmission [9]. At the transport level, application data is augmented by headers and subject to retransmissions. In poor infrastructure conditions, retransmissions can greatly increase the trafc exchanged. The authors showed that mobile billing policies vary; some charge users for retransmissions whilst others allow free-riding through abuse of transport protocols. By highlighting potential attacks, and suggesting mitigation techniques, the authors hope that future billing techniques might accurately reect a users application use. The aspiration towards perfect accounting attracted attention during audience questions the suggestion of unlimited data plans as an alternative to accurate billing was made. Thad Starner also highlighted the energy cost of retransmissions and the opportunity for such retransmission attacks to potentially jam an already limited mobile data service.
2

The session (chaired by Nigel Davies, Lancaster University) opened with How is Energy Consumed in Smartphone Display Applications? [6] presented by Mian Dong (Samsung Telecommunication America). The work explored the power consumption of smartphone AMOLED screens through two research questions: the rst focused on improvements to AMOLEDs over time and found that the practical power improvement is relatively low. The second question focused on producing a power analysis for the display in an application context. For three sample applications (video, gameplay, camera), the display was found to be a relatively small contributor to power consumption. However, some content can result in high power consumption, which may be reduced using colour transformation (e.g., reducing brightness, increasing saturation). Mostafa Uddin (Old Dominion University) then presented A2PSM: Audio Assisted Wi-Fi Power Saving Mechanism for Smart Devices [16]. Wi-Fi power saving mechanisms (PSMs) are key for extending the limited power resources of mobile devices. A2PSM exploits lower-powered audio hardware to improve existing PSMs by reducing the need for wake periods. Evaluation of a smartphone prototype suggests A2PSM could offer in excess of 25% improvement over static PSM with no impact on network throughput. The nal paper of the session, Application Modes: A Narrow Interface for End- User Power Management in Mobile Devices [11] was presented by Marcelo Martins (Brown University). Application Modes are suggested as a mechanism for allowing users to make power management decisions on their mobile device. Each application can offer a number of modes that comprise of reductions of functionality with associated power savings (e.g., turn off automatic synchronisation). Modes allow developers to provide graceful degradation of their application in resourceconstrained circumstances, and also allow the user to prioritise an in-use application over background applications. At the close of this session, discussion focused around the idea that mobile computing has potentially reached a point where all easy optimisations are gone; further energy improvements must impact the user. Ofoading decisions to the user is difcult and it is difcult to know which applications are draining battery. Motivating developers to conserve energy, and raising awareness of application energy use, could be

Mobile Computing and Communications Review, Volume 17, Number 2, April 2013

targeted in future work.

VI.

Understanding Humans

from Duke. The main argument laid on whether users would be willing to expose themselves in exchange of functionality and what the system could do with the collected information about users.

The session, chaired by Anthony LaMarca (Intel Labs), opened with Xuan Bao (Duke University) presenting The Case for Psychological Computing [1]. Bao advocated a system capable of modelling human psychological attributes as inherent nature (longterm, xed values) and states (short-term, contextdependent values). He believes that developers can make use of these variables to create applications that better meet human properties, e.g., preferential content pre-fetching and vehicle driving with cognitive comfort. Bao concluded his talk by mentioning some of the remaining challenges for enabling psychological computing. These include how to model attributes correctly, and in case of wrong prediction, how to back off gracefully. He Wang (Duke University) followed with InSight: Recognizing Humans without Face Recognition [17], a presentation of a system capable of identifying humans using camera-enabled portable devices in situations where computer vision cannot distinguish face patterns. Wang found that spectogram analysis applied to cloth colors and wavelets applied to cloth patterns are prominent techniques for accurately recognising humans from their garment. InSight can share these ngerprints with other peoples devices to enable social-aware applications. Shahriar Nirjon (University of Virgina) presented the last paper of the session, sMFCC: Exploiting Sparseness in Speech for Fast Acoustic Feature Extraction on Mobile Devices a Feasibility Study [12]. Nirjon noted that current solutions for voice-driven smartphone applications must ofoad the extraction of smartphone audio features to the cloud to guarantee real-time system response, which leads to high communication cost. After empirically observing the sparseness of frequency domain in speech, Nirjon proposed sMFCC (sparse Mel-Frequency Cepstral Coefcients) to exploit this property and enable in-situ, fast acoustic feature extraction without relying on the cloud. Nirjon built sMFCC on top of recent work on sparse Fast Fourier Transform (sFFT) as an efcient approximation to the widely used MFCC algorithm. He showed that sMFCC can achieve an accuracy of up to 83.97% on word recognition with a low expected response time. Privacy was the key topic during the discussion session, raising concerns on unwanted disclosure of personal data, especially in the case of both presentations

VII.

Sensors and Data

Rajesh Krishna Balan (Singapore Management University) chaired the session. Yu Xiao (CMU/Aalto University) started her talk, Lowering the Barriers to Large-Scale Mobile Crowdsensing [19], by listing the main obstacles to a successful scaling of collaborative sensing on mobile devices: device heterogeneity, lack of user incentives and high communication cost. She urged the community to rethink the deployment model for this type of applications and suggested a three-tier architecture to ease scaling. She proposed the use of cloudlets as a middle layer between mobile clients and app server. Cloudlets work as a distributed infrastructure for running VMs, physically close to the mobile client to reduce networking costs. Cloudlets comprise of a set of proxy VMs, which abstract device heterogeneity and handle requests of sensor data on behalf of mobile clients, and application VMs, which decentralise application logic and data aggregation. She explained that the app server manages the transfer of VMs between cloudlets according to the users position, in order to guarantee low-latency logic ofoading. She concluded her presentation posing some technical challenges to enabling this new technology virtualisation overhead and cloudlet reconguration due to VM migration. Waylon Brunette (University of Washington) followed with Open Data Kit 2.0: Expanding and Rening Information Services for Developing Regions [3]. Brunette elaborated on the success of Open Data Kit 3 (ODK) 1.0 as a modular toolkit for building mobile applications mainly used for data collection and aggregation in developing regions. From the feedback received from various developers worldwide, Brunette iterated over the drawbacks of ODK 1.0 and completed his talk by explaining how his team at UW xed these shortcomings in ODK 2.0, enabling the creation of more complex applications. Supriyo Chakraborty (UCLA) concluded the session with A Framework for Context-Aware Privacy of Sensor Data on Mobile Devices [5], a study on the tradeoff between application utility and shared-data privacy. Chakraborty argued about the risk of disclosing sensitive information by untrusted apps via inferences on shared personal data. To preserve anonymity
3

http://opendatakit.org
3

Mobile Computing and Communications Review, Volume 17, Number 2, April 2013

and avoid private-data reconstruction, the presenter hinted on converting data to feature spaces with lower dimensions before sharing and partially obfuscating these features by sharing only what is explicitly vetted by the user. Chakraborty nished his talk by describing ipShield, an inference privacy framework implemented atop Android that models the adversary capability of inferring sensitive information from shared data, and provides a rewall for coping with datasharing lters for apps and data obfuscation rules by users. A major part of the concluding discussion centered on Chakrabortys work and the implications of privacy to the user. The audience raised questions about ipShields maintenance, as it requires expert knowledge to update lters and rules, which seems impractical to users. A debate followed on which simple principles of privacy could be designed to reduce this reliance on specialists for particular applications.

P2P system that his team at MSR designed to share information downloaded from the Web with peers when there is very limited connectivity. The discussion following the panelists speeches focused on the scope and feasibility of ICTD as a research area. The panelists noted that the IT disparage not only exists because of economical problems, but also due to cultural issues that affect technological absorption. They also noted that community efforts should focus on problems that will linger for many generations. Finally, the panelists declared that research on ICTD is a gratifying experience, as they create opportunities for local training and local employment by start-up companies.

IX.

Mobile Cloud Interactions

VIII.

Panel Mobile Systems and the Developing World

Gaetano Borriello (University of Washington) moderated the panel on Information and Communication Technology for Development (ICTD) applied to mobile computing. Recent developments in mobile computing and communication led to the proliferation of mobile phones worldwide. Thanks to their lower prices, these devices have become the main platform access to technology in emerging markets, substituting notebooks and desktop computers. Three academic experts provided their input based on their research experience. Elizabeth Belding (UCSB) summarised some of her technological partnerships with African communities to bring wireless Internet connectivity and services to small villages. Belding introduced her work with UCSB students on devising mechanisms for sharing trafc and content using wireless technology and reducing costs of cellular and data telephony in remote locations. Lakshmi Subramanian (New York University) gave the audience a glimpse of his research on mobile solutions to bridge the digital divide: a web search engine based on SMS input, keeping connectivity in unstable rural cellular networks, among others. Subramanians attention and adaptation to resource constraints was particularly impressive. Bill Thies (MSR India) continued the series of design for extremes by showing some of his creativity to bring research opportunities using new solutions for constrained environments. In special, Thies showed a
4

The session, chaired by Matt Welsh (Google), opened with Lars Tiede (University of Troms) presenting Cloud Displays for Mobile Users in a Display Cloud [15]. Tiede introduced the cloud display, an abstraction for public, shared programmable displays which can be discovered, composed and congured by mobile devices. Visual output comes from applications running on desktop computers, whilst mobile devices serve as advanced remote controllers. Tiede demonstrated a functioning prototype based on VNC technology for screen arrangement, authentication, content display and transfer. He showed that the overhead generated from the communication between displays and app-running device is negligible considering the current network technology. Jeffrey Bickford (AT&T Security Research Center) followed with Towards Synchronization of Live Virtual Machines among Mobile Devices [2]. His presentation focused on the design of VMsync, a service to enable users to switch seamlessly between mobile devices with both data and computation states preserved. He explained that VMsync aims for minimal user-perceived delay during switching by incrementally transferring checkpoints from one active VM on one device to other devices. For the rest of his talk, Bickford analysed different synchronisation policies for representative multimedia mobile workloads. He observed that sending deltas between checkpoints decreases the synchronisation latency, and although increasing the checkpoint interval can help reduce memory and le-system delta sizes, such approach may risk the promised seamlessness in user experience. Michael Butkiewicz (UC Riverside) completed the roster with Enabling the Transition to the Mobile Web with WebSieve [4], a study on optimising the

Mobile Computing and Communications Review, Volume 17, Number 2, April 2013

load time of websites in mobile devices. Butkiewicz based his work on the observation that a great number of popular websites do not provide an equivalent mobile-friendly version, resulting in high load times and user-experience deterioration. He proposed WebSieve, an architecture to estimate the user-perceived value of web objects on a page, and, given a loadingtime constraint, only load those objects that have the highest value to the user. Based on user surveys, Butkiewicz discovered that different users perceive different utilities for the same set of objects, hence WebSieve necessitates personalisation. Finally, he described his prototype design as an optimisationproblem solver over a dependency graph and covered some of the non-trivial research questions WebSieve brings about. The discussion session started with a hot debate between advocates and skeptics of the applicability and usefulness of programmable displays. While the supporters pointed to public scenarios where displays can be useful, a few contenders still speculate on the killer app. The second part of the discussion concentrated on the challenges raised by WebSieve as a scalable solution for different websites. Butkiewicz pointed out that it is still early to draw any conclusions, but he has already collected promising results from his prototype implementation.

application which allows drivers to post or listen to voice tweets grouped by a vehicular social network (VSN, a social network of drivers who travel the same routes or to the same destination). Voice tweets are formed into digests by the NaviTweet server and then sent to drivers mobile device for playback. Whilst the audio nature of the tweets was intended to reduce cognitive overload, audience questions highlighted the potential for background noise to make interpretation challenging and even for the opportunity for transmission of negative, road rage messages. The nal paper, Quantifying the Potential of RideSharing using Call Description Records [7], was presented by Blerim Cici (UC Irvine). The work attempts to quantify the upper bound on ride-sharing by considering human mobility patterns using a 3-month call description record (CDR) dataset. By permitting short detours to pickup/drop-off, analysis shows that ride-sharing has the potential to eliminate 50% of car journeys. During discussion following Cicis presentation, audience members considered methods of encouraging ride-sharing by identifying incentives to share (e.g., through use of social graphs) and increasing opportunity by allowing passengers divide their route between cars. The potential to improve public transportation through CDR analysis was also suggested.

X.

Vehicular Networking and Transportation

XI.

Posters & Demos

The nal session, chaired by Lin Zhong, opened with the paper Scout: An Asymmetric Vehicular Network Design over TV Whitespaces [20] presented by Tan Zhang (University of Wisconsin-Madison). Motivated by an increasing trend towards in-vehicle Internet connectivity, Scout provides an architecture for vehicular networking by leveraging locally available television frequencies (whitespaces). Power transmission restrictions suggest an asymmetric design so whitespace is used as a downlink whilst cellular 3G is used for the uplink. Two radios are used per vehicle, the front (scout) monitors changing channel characteristics (e.g., the vehicle passing behind a building) and informs the base station which then optimises communication with the rear radio. Initial evaluation indicates that Scout can improve coverage by 4 and increase throughput by 1.4. Wenjie Sha (Rutgers University) then presented Social Vehicle Navigation: Integrating Shared Driving Experience into Vehicle Navigation [14]. Sha described their system NaviTweet, a social navigation

Winner of the best poster/demo award, Bringing Insitu Social Awareness to Mobile Systems: Conversational Turn Monitoring and its Applications [10] (Lee et al., KAIST) attracted signicant attention and discussion from attendees. Their system uses the smartphone as a mechanism for encouraging inperson interactions by monitoring conversation patterns. The demo was supported by a video of the system in use in a nursery encouraging peerengagement from an otherwise withdrawn autistic child. By combining the video with a layout of smartphones running their turn-monitoring system (which used coloured bars to indicate speakers and conversation duration), attendees were able to see a clear demonstration of the capability of the system as well as potential applications and were rewarded for their interactions with the demonstrators with a clear representation of their conversation patterns. Leveraging Imperfections of Sensor for Fingerprinting Smartphones [8] (Dey et al., University of South Carolina) received an honourable mention as the second best poster/demo of ACM HotMo5

Mobile Computing and Communications Review, Volume 17, Number 2, April 2013

bile13. Their work suggests uniquely identifying smartphones by comparing the responses from a given stimulus applied to a built-in sensor chip of different devices. As imperfections from manufacturing these chips can subtlety affect their reading responses, sensor ngerprinting seems to be a promising alternative for telling devices apart, especially considering that smartphone IMEI codes can be duplicated. Wee at al.s demo (Singapore Management University), Focus: A Usable & Effective Approach to OLED Display Power Management [18] received the award for third best poster/demo. Following on from some of the power management discussions, this demo showed how power saving can be achieved by powering off sections of the screen that do not need to be displayed. Conference attendees could create their own dimming prole for one of a number of predened Android applications. Their demo allowed attendees to explore the usability of an application with dimmed portions (most often the unused lower portions of the screen) and to see the potential battery-life gains.

[5] S. Chakraborty, K. R. Raghavan, M. P. Johnson, and M. B. Srivastava. A framework for context-aware privacy of sensor data on mobile systems. In ACM HotMobile, 2013. [6] X. Chen, Y. Chen, Z. Ma, and F. C. A. Fernandes. How is energy consumed in smartphone display applications? In ACM HotMobile, 2013. [7] B. Cici, A. Markopoulou, E. Fras-Martnez, and N. Laoutaris. Quantifying the potential of ridesharing using call description records. In ACM HotMobile, 2013. [8] S. Dey, N. Roy, W. Xu, and S. Nelakuditi. Leveraging imperfections of sensors for ngerprinting smartphones. In ACM HotMobile, 2013. [9] Y. Go, D. F. Kune, K. Park, and Y. Kim. Towards accurate accounting of cellular data for TCP retransmission. In ACM HotMobile, 2013. [10] Y. Lee, C. Min, C. Hwang, J. Lee, I. Hwang, C. Yoo, and J. Song. Bringing in-situ social awareness to mobile systems: Conversational turn monitoring and its applications. In ACM HotMobile, 2013. [11] M. Martins and R. Fonseca. Application modes: a narrow interface for end-user power management in mobile devices. In ACM HotMobile, 2013. [12] S. Nirjon, R. Dickerson, J. Stankovic, G. Shen, and X. Jiang. sMFCC: exploiting sparseness in speech for fast acoustic feature extraction on mobile devices a feasibility study. In ACM HotMobile, 2013. [13] H. Raj, S. Saroiu, A. Wolman, and J. Padhye. Splitting the bill for mobile data with SIMlets. In ACM HotMobile, 2013. [14] W. Sha, D. Kwak, B. Nath, and L. Iftode. Social vehicle navigation: integrating shared driving experience into vehicle navigation. In ACM HotMobile, 2013. [15] L. Tiede, J. M. Bjrndalen, and O. J. Anshus. Cloud displays for mobile users in a display cloud. In ACM HotMobile, 2013. [16] M. Uddin and T. Nadeem. A2PSM: audio assisted wi- power saving mechanism for smart devices. In ACM HotMobile, 2013. [17] H. Wang, X. Bao, R. R. Choudhury, and S. Nelakuditi. InSight: recognizing humans without face recognition. In ACM HotMobile, 2013. [18] T. K. Wee, T. Okoshi, A. Misra, and R. K. Balan. DEMO of Focus: A usable & effective approach to OLED display power management. In ACM HotMobile, 2013. [19] Y. Xiao, P. Simoens, P. Pillai, K. Ha, and M. Satyanarayanan. Lowering the barriers to large-scale mobile crowdsensing. In ACM HotMobile, 2013. [20] T. Zhang, S. Sen, and S. Banerjee. Scout: an asymmetric vehicular network design over TV whitespaces. In ACM HotMobile, 2013.

XII.

Concluding Remarks

ACM HotMobiles size and interactive nature makes it a unique platform for the discussion of mobile systems, applications and technologies. The state-ofthe-art research presentations and demonstrations are highly interactive and engaging.

Acknowledgments
The authors gratefully acknowledge nancial support in attending ACM HotMobile 2013 through the ACM HotMobile travel grant scheme.

References
[1] X. Bao, M. Gowda, R. Mahajan, and R. R. Choudhury. The case for psychological computing. In ACM HotMobile, 2013. [2] J. Bickford and R. Cceres. Towards synchronization of live virtual machines among mobile devices. In ACM HotMobile, 2013. [3] W. Brunette, M. Sundt, N. Dell, R. Chaudhri, N. Breit, and G. Borriello. Open data kit 2.0: expanding and rening information services for developing regions. In ACM HotMobile, 2013. [4] M. Butkiewicz, Z. Wu, S. Li, P. Murali, V. Hristidis, H. V. Madhyastha, and V. Sekar. Enabling the transition to the mobile web with WebSieve. In ACM HotMobile, 2013.
6

Mobile Computing and Communications Review, Volume 17, Number 2, April 2013