GUEST EDITORS’ INTRODUCTION

Pervasive Computing Revisited

n 2001, Bahl, Gaetano Borriello, Nigel Davies, Hans (Satya) published a paper, “Pervasive Gellersen, Jason Hong, Liviu Iftode, Anupam Computing: Vision and Challenges,” Joshi, David Kotz, James Landay, Marc Lang- in IEEE Personal Communications.1 heinrich, Dejan Milojicic, and Dan Siewiorek. This was the first paper to substantively It has also inspired at least nine papers with Iexplore the implications of ’s 1991 more than 500 citations each (see the “Inspiring vision for a real-world systems infrastructure.2 Others: Related Work” sidebar). It provided the conceptual foundations for what Looking back at the vision Satya outlined much of the computing indus- and comparing it to today’s reality, we see that try now calls the Internet of our community has made progress, but we still Maria R. Ebling Things (IoT). It also developed have a long way to go. For example, in two sce- IBM T.J. Watson Research Center a taxonomy showing the deep narios he outlines, we can see aspects of every- Roy Want relationships between distrib- day pervasive computing—we use cloud storage uted systems, mobile comput- services that automatically sync our electronic ing, and pervasive computing documents between devices, and we identify (IoT). You might not realize which documents to maintain on each of our it, but the paper also inspired devices. But many aspects of his scenarios are the creation of our own publication, IEEE Per- more distant. For example, my has vasive Computing, with Satya serving as the never told me to move to a different area to get founding Editor in Chief. better network service. Last year, the IEEE Pervasive Computing A Look Back editorial board decided that after slightly more Satya’s paper has earned almost 3,000 cita- than 15 years, this would be a good time to look tions and has inspired many well-known au- back at Satya’s vision, review the challenges thors whose names are synonymous with the he laid out for our community, and evaluate pervasive computing field, including Victor what has been achieved in making the vision a

Published by the IEEE CS n 1536-1268/17/$33.00 © 2017 IEEE PERVASIVE computing 17 GUEST EDITORS’ INTRODUCTION

Inspiring Others: Related Work

he following important works were all inspired by Satya’s • J.I. Hong and J.A. Landay, “An Architecture for Privacy- T 2001 paper, “Pervasive Computing: Vision and Sensitive ,” Proc. 2nd Int’l Conf. Mobile Challenges,” in IEEE Personal Communication: Systems, Applications, and Services (MobiSys), 2004, pp. 177– 189; http://dl.acm.org/citation.cfm?id5990 087. • D.S. Milojicic et al., Peer-to-Peer Computing, tech. report • D. Cook and S.K. Das, Smart Environments: Technology, Proto- HPL-2002-57 (R.1), HP Labs, 2002; www.hpl.hp.com/techre- cols and Applications, John Wiley & Sons, 2004. ports/2002/HPL-2002-57R1.pdf. • L. Kagal, T. Finin, and A. Joshi, “A Policy Language for a Per- • D. Garlan et al., “Project Aura: Toward Distraction-Free vasive Computing Environment,” Proc. IEEE 4th Int’l Workshop Pervasive Computing,” IEEE Pervasive Computing, vol. 1, no. Policies for Distributed Systems and Networks, 2003; http:// 2, 2002, pp. 22–31; http://ieeexplore.ieee.org/abstract/ ieeexplore.ieee.org/abstract/document/1206958/. document/1012334. • D.J. Cook and S.K. Das, “How Smart Are Our Environments? • D. Saha and A. Mukherjee, “Pervasive Computing: A Paradigm An Updated Look at the State of the Art,” Pervasive and Mo- for the 21st Century,” Computer, vol. 36, no. 3, 2003, pp. 25–31; bile Computing, vol. 3, no. 2, 2007; https://doi.org/10.1016/j. http://ieeexplore.ieee.org/abstract/document/1185214. pmcj.2006.12.001 • M. Pantic and L.J.M. Rothkrantz, “Toward an Affect-Sensitive • E. Aarts, J. Korst, and W.F.J. Verhaegh, “Algorithms Multimodal Human-Computer Interaction,” Proc. IEEE, vol. in Ambient Intelligence,” Phillips Research Book Se- 91, no. 9, 2003, pp. 1370–1390; http://ieeexplore.ieee.org/ ries, vol. 2, Springer, 2004; https://link.springer.com/ abstract/document/1230215. chapter/10.1007/978-94-017-0703-9_1.

reality. This special issue is a retrospec- how cyber-foraging systems might In this article, he categorizes pervasive tive on that vision. reach mainstream deployment. devices into three tiers: the top tier (tab- In “Powering Pervasive Comput- lets, , and so on), middle In This Issue ing Systems,” Shwetak Patel and tier (TVs, refrigerators, thermostats, We open with an interview with Satya Joshua Smith examine the progress and so on), and bottom tier (including in which he looks back at the challenges made in power delivery and manage- RFID tags; sensors; and heating, ven- he identified in 2001, noting what he ment over the past 15 years, discuss- tilation, and air conditioning systems). missed and highlighting what has been ing today’s trends in advancing this He then describes six types of privacy more difficult than expected. The inter- important area. They examine power- concerns: the ease with which people view touches on privacy, smart spaces, harvesting approaches as well as power are aware of what is being sensed, the and user distraction and intent. It also transfer through both near-field and time scope of the sensing, the number explores the relationship between per- far-field mechanisms. They also point of input and output capabilities the vasive computing and both IoT and out the need to architect pervasive sensing technology can use to interact . We think you will computing systems that recognize and with users, the amount of software find Satya’s observations insightful. embrace the power limitations of dif- that can run on the device, the exis- One of the key ideas identified in Satya’s ferent technologies while minimizing tence of third-party software that can seminal paper is cyber foraging—he the need for power at the most energy- be installed and can run on the device, argued that future mobile devices constrained layers of the system. They and the amount of support manufac- would have to acquire the resources also point out that we have biological turers provide in terms of fixing bugs they need from the environment examples of systems that are millions and upgrading software. Hong then around them. In “Cyber Foraging: of times more energy efficient than cur- considers each of these dimensions Fifteen Years Later,” Rajesh Krishna rent technologies, so they’re optimistic across each device tier, giving readers Balan and Jason Flinn examine the about the outlook for continued ad- a good sense of the range of concerns progress that has been made in this vances in energy efficiency. that must be considered when build- area. In particular, they discuss seam- Jason Hong shares his thoughts ing a new device or installing a device less cyber foraging and look at what about privacy in pervasive computing in a home. technical challenges remain, how those systems in his article, “The Privacy Finally, in “Pervasive Data Science,” gaps hamper adoption, and when and Landscape of Pervasive Computing.” Nigel Davies and Sarah Clinch make

18 PERVASIVE computing www.computer.org/pervasive the AUTHORS the case that the intersection between Maria R. Ebling is a director at the IBM T.J. Watson Research Center. Her pervasive computing and data science research interests include pervasive computing, context-aware computing, entails some unique challenges that must mobile computing, distributed systems, privacy, and human-computer inter- action. Ebling received her PhD in from Carnegie Mellon be addressed. Examples of applications University. Contact her at [email protected]. at this intersection, which they discuss, include augmented cognition to improve human abilities, autonomous vehicles to transform our transportation infra- Roy Want is a research scientist at Google. His research interests include mo- structure, and smart spaces to improve bile and ubiquitous computing. Want received his PhD from Uni- our physical environment. They out- versity. He is an ACM and IEEE Fellow. Contact him at [email protected]. line five challenges specific to pervasive data and cover the entire data pipeline. These challenges, like those of pervasive computing more broadly, will require an interdisciplinary approach. The article walks readers through two use cases demonstrating the end-to-end challenges technologies converge with research REFERENCES of pervasive data science. Ultimately, progress on the many fundamental Davies and Clinch argue that pervasive problems discussed in this paper,” was 1. M. Satyanarayanan, “Pervasive Com- puting: Vision and Challenges,” IEEE data science will transform our ability to most certainly true. We are well into Personal Comm., vol. 8, no. 4, 2001, pp. achieve the goals of ubiquitous comput- the second decade of the 21st century, 10 –17. ing outlined by Weiser. Readers looking and we continue to see many research 2. M. Weiser, “The Computer for the 21st Cen- for a thesis topic should find the chal- initiatives designed to solve the chal- tury,” Scientific Am., Sept. 1991, pp. 66–75. lenges presented in this paper inspiring. lenges that Satya laid out for us in this one, short article. It’s always amazing to atya’s prediction that the “early see how Satya can keep so many people Read your subscriptions decades of the 21st century busy solving difficult and meaty prob- through the myCS will be a period of excitement lems, but he seems to do it again and publications portal at http://mycs.computer.org. S and ferment, as new hardware again! Call for Articles

IEEE Pervasive Computing

seeks accessible, useful papers on the latest

peer-reviewed developments in pervasive,

mobile, and ubiquitous computing. Topics

include hardware technology, software

infrastructure, real-world sensing and

Author guidelines: interaction, human-computer interaction, www.computer.org/mc/ and systems considerations, including pervasive/author.htm deployment, scalability, security, and privacy. Further details:

[email protected] www.computer.org/pervasive

JULY–SEPTEMBER 2017 PERVASIVE computing 19 Interview

Satya Revisits “Pervasive Computing: Vision and Challenges”

Maria R. Ebling, IBM T.J. Watson Research Center Roy Want, Google

n this interview with Mahadev Saty- 1991 and 2001 had been hugely trans- Ianarayanan (Satya), the founding formative in computing, with the emer- Editor in Chief of IEEE Pervasive gence of the World Wide Web and public Computing, we learn about the moti- awareness of the Internet, the total dom- vation for his 2001 paper, “Pervasive inance of personal computing and the Computing: Vision and Challenges” disappearance of timesharing, the dire (IEEE Personal Communications, vol. fates of flagship companies—such as 8, no. 4). Satya also considers here how the disappearance of Digital Equipment much progress we have made and what Corporation and the near-death experi- new challenges have since emerged. ence of IBM—the Dot-Com boom, the Satya has been a leader in the field end of the Cold War and its impact on of pervasive computing for many years, DARPA’s contribution to computing pioneering advances in distributed innovations, and so on. In technical and systems, mobile computing, and the non-technical aspects, the computing . He is the Carnegie landscape of 2001 was very different Group Professor of Computer Science from what it was in 1991. In writing at Carnegie Mellon University, and he the paper, I was revisiting Weiser’s high- is a Fellow of the ACM and IEEE. He level vision in the light of all that had was the founding program chair of the far-out, challenging, exciting, and high- happened and all that we had learned in HotMobile series of workshops, the impact that excellent computer science that turbulent decade. founding program chair of the IEEE/ researchers should engage deeply in Looking back, it is fair to say that ACM series of conferences on Edge it. Mark Weiser, of course, said it first Weiser’s vision has lived on and con- Computing, and the founding director and most beautifully, which is why the tinues to inspire excellent research. of Research Pittsburgh. paper begins with a quote from his To the extent that my 2001 paper has 1991 Scientific American paper, “The helped to sustain and extend Weiser’s It has been 15 years since you published Computer for the 21st Century.” With vision, it has been a success. I am grate- “Pervasive Computing: Vision and Weiser’s passing in 1999, and with the ful to have been at the right place, at Challenges.” Looking back, what moti- Xerox PARC lab he created no longer the right time, to write it. The creation vated you to write the paper, and what leading the charge, I felt that his vision of IEEE Pervasive Computing was the aspects of the vision were spot on? was unlikely to be pursued further most direct consequence of writing Certainly, the high-level prediction unless the research community con- the paper, because it made clear what that this whole area of research would sciously dedicated itself to the quest. a rich set of open questions lay before be very active, fruitful, and long-lived Weiser’s vision was expressed at a very us. That helped to bring together and has proven true. That was the real point high level, while my 2001 paper tried to create the community consensus and of the paper—to make the case that zoom in on some key technical chal- enthusiasm that lead to the found- the high-level vision was sufficiently lenges. In addition, the decade between ing editorial board. It also helped the

20 PERVASIVE computing Published by the IEEE CS n 1536-1268/17/$33.00 © 2017 IEEE IEEE Computer Society see this as a content pertaining to advertising, and Weiser placed the human at the center viable initiative. Looking back, it is the many round trips to load a typi- of his vision. Making sensing, com- impressive to see all the high-quality cal deeply nested webpage today, both puting, and communication technol- work that has been published in IEEE conspire to increase network use and ogy “disappear” was his holy grail. In Pervasive Computing since its found- hence energy demand. These devel- contrast, the IoT vision is silent about ing, as well as the many conferences, opments have muted the benefits of the role of humans. It would therefore workshops, successful funding propos- many improvements in energy demand not be a stretch to say that pervasive als, PhD and master’s theses, and so on in other aspects of mobile hardware computing subsumes IoT—it covers all that have emerged in pervasive comput- and software. On the supply side, the that IoT covers and more. ing over the past 15 years. The number of researchers who have been actively It would not be a stretch to Back in 2001, cloud computing was engaged in this research is quite large, not a popular concept. Were you sur- compared to where it was in 2001. say that pervasive computing prised by the later dominance of this Today, this content is a key part of subsumes IoT—it covers all model, and did it address any of the many graduate courses worldwide. that IoT covers and more. challenges you presented? Indeed, cloud computing was not Do you see any challenges that were on anyone’s radar in 2001—except, of not obvious at that time? energy density (joules per kilogram) course, at VMware. Its emergence was One thing that I was not thinking of batteries is improving at a glacial unrelated to challenges in mobile and about in 2001 was how to sustain com- pace. As early as 1997, it was noted in pervasive computing and was driven mercial products in this space. Today, an NRC report that the energy density mainly by factors pertaining to the cost advertising revenue is the foundation of of lithium-ion batteries was half that of of owning and operating computing many business models that relate to this dynamite—the only difference is the infrastructure, such as enterprise data space. Unfortunately, by its very nature, rate of discharge. The recent anecdotes centers. However, once cloud comput- advertising is about consuming spare user of exploding lithium-ion batteries rein- ing emerged, it was rapidly leveraged attention. It therefore runs counter to the force this point. So although I did fore- in mobile and pervasive computing. In concept of lowering distraction, which see battery life as a challenge, it is fair fact, one of its early uses was for cyber was the core of Weiser’s vision for ubiqui- to say that the challenge has been even foraging—specifically, for offloading tous computing. How to square this circle more difficult than anticipated. compute-intensive operations from a is a difficult challenge for which I have not Other challenges that have proven mobile device. Apple’s Siri speech rec- seen any good solutions yet. more difficult than anticipated include ognition system appeared soon after the measuring user distraction and infer- release of the iPhone in 2007. The voice Were any of the challenges you saw ring user intent. signal was captured on a smartphone, more difficult than you originally sent to the cloud for processing, and the thought? In recent years, there has been great result was returned to the mobile user. A number of the challenges men- interest in the Internet of Things. How Many other applications that have since tioned in the paper continue to be do you view the relationship between appeared—including Google Voice and difficult. We have made progress on pervasive computing and IoT? Google Goggles—use a similar strategy. them but still have a long way to go, It is the same thing, with a different On the one hand, the appearance of and in some cases new complica- name and slightly different focus. The the cloud simplified the answer to the tions have arisen to make the problem essence of the pervasive computing question of where offloading should be harder. Consider battery life. In spite of vision is a world saturated with sensing done. At the same time, using the cloud advances at many levels over the last 15 and computing, yet so gracefully inte- implied large end-to-end latency. This years, the battery life of mobile devices grated with humans that all this technol- is suboptimal for important use cases, is still a big concern. Experiments with ogy remains below their consciousness. and it led to the concept of cloudlets. Google Glass show a battery life of at The essence of the IoT vision is a A separate use of the cloud has been most an hour when it is being put to world full of intelligent devices that the creation of “hubs” to collect data serious use. If you use GPS on your sense the physical world and integrate from sensors at the edge of the Inter- smartphone for navigation, the bat- their observations to meet some higher net—examples include home thermo- tery typically lasts much less than a full level goal (such as the energy efficiency stats and water flow meters. However, day. The growing number of sensors on of a building or early failure warning cloud-based IoT hubs pose scalability mobile devices, combined with many of a jet engine). In terms of technical challenges for high-data-rate sensors asynchronous network activities, have content relating to sensing, computing, such as video cameras. They also pose increased energy demand significantly. and communication, these are virtually privacy challenges, because users don’t The reduction in the effectiveness of identical visions. The one substantial control the data—users are not given caching, mainly due to dynamic web difference lies in the role of the human. the opportunity to redact or denature

JULY–SEPTEMBER 2017 PERVASIVE computing 21 INTERVIEW INTERVIEW

the data before it is released. These con- of dispersed elements, with no hard One way to think of this is that mobile cerns are also addressed by the use of state, called cloudlets. We described devices can be used to make a space cloudlets. the cloudlet concept in the October– temporarily smart. So smart spaces do December 2009 issue of IEEE Perva- exist, but they are being created tran- How does your more recent cloudlet sive Computing in “The Case for VM- siently through the use of smart mobile proposals to support edge computing Based Cloudlets in Mobile Computing,” devices. In general, computer vision play into the pervasive vision, then and which has proven highly influential and using a combination of mobile devices now? has been viewed as the founding mani- and associated cloudlets can be used to Implicit in the cyber foraging meta- festo of edge computing. Cisco’s more make ordinary spaces “smart.” phor is the notion of “nearby” resources. recent concept of “fog computing” is It seemed obvious to me when writing essentially the same concept. Will the implementation of physical the 2001 paper that low latency and high smart spaces be entirely replaced by bandwidth to the remote execution site One of the key challenges you identi- transient smart spaces, or is there still were essential attributes, so I never men- fied focused on smart spaces. What do a place for physical smart spaces? tioned them explicitly in the paper. Prox- you think is the most successful adop- The ability to transform any space imity is thus not explicitly mentioned in tion of smart spaces, and why haven’t temporarily into a smart space is a the discussion of cyber foraging. such spaces become more common? powerful capability. A collection of co- The emergence of cloud comput- This is an excellent question. Mobile located mobile users could bring a smart ing circa 2008 simplified some things devices are definitely much smarter space into existence wherever they hap- related to cyber foraging. The view of today, but it is fair to say that there pen to be. I think we will see more ways the cloud as a single, logically central- are fewer smart spaces than we antici- of making ordinary spaces smart in this ized and managed entity that serves as pated in 2001. One reason perhaps is way. For example, at MobiSys 2014, Jun- a computing utility simplified questions the economics of smart spaces. It takes jue Wang and his colleagues presented relating to resource discovery, trust, substantial investment by an organiza- “Ubiquitous Keyboard for Small Mobile and business models. “In the cloud” tion to create a smart space, such as a Devices: Harnessing Multipath Fading was the obvious answer to the ques- conference room that is richly equipped for Fine-Grained Keystroke Localiza- tion, “Where should remote execution with sensors and actuators. This cost is tion.” They discussed their work on be performed?” typically borne by the organization. projecting a keyboard onto any flat Unfortunately, the global consolida- The level of technology is frozen for surface for use as a normal keyboard. tion of cloud computing implies large some time after the creation of a smart And at a 2015 Lenovo Techworld event, average separation between a mobile space. It is difficult and expensive to Lenovo presented prototype hardware device and its cloud. End-to-end com- continuously track improvements in with the ability to project directly from munication then involves many network smart space technology. In contrast, the a smartphone onto a wall—essentially hops and results in high latencies. By cost of mobile devices is typically borne a “picoprojector” [www.youtube.com/ early 2008, I realized that my implicit by individual users. If a user carries a watch?v=JwBem1Ul8dk]. The proto- assumption of “nearby” in framing the smartphone and wears a FitBit, then he type can also project a keyboard onto cyber foraging concept was a mistake. or she pays for them both. a flat surface. I should have made explicit the impor- An excellent example of this eco- The digital camera replacing a smart tance of proximity. nomic tradeoff is the use of smart whiteboard is another example. So I do Discussions with a number of senior whiteboards. In 2001, there were many believe that “on the fly” smart spaces researchers in mobile computing at the companies working on the creation of are going to become increasingly impor- 2008 MobiSys conference in Breck- smart whiteboards that could capture tant. Many years ago, at HotMobile enridge, Colorado convinced me that what was written or drawn on them. 2008, Roy Want articulated the vision it was necessary to make the case for However, the proliferation of smart- of “dynamic composable computing.” proximity explicit to the research com- phones with digital cameras has made In a way, this is a generalization of that munity and to industry. In close collab- the smart whiteboard concept obsolete. concept to larger physical spaces. oration, Victor Bahl from , It is trivial to take a picture of a white- That said, there will always be a place Roy Want from Intel, Ramón Cáceres board with your smartphone before you for physical smart spaces. For example, from AT&T, Nigel Davies from Lan- erase it. Emailing those pictures to all there might be security reasons why caster University, and I articulated the participants is trivial. Optical character certain discussions should only occur need for a two-level architecture for recognition of the images in the cloud within the confines of a controlled mobile-cloud convergence. The first can produce text output. A smart space space. There might also be special- level is today’s unmodified cloud infra- component has thus been displaced by ized hardware capabilities that are structure. The second level consists a smart mobile device. not portable. For example, the Icaros

22 PERVASIVE computing www.computer.org/pervasive INTERVIEW

harness [www.icaros.net] makes you other sensors to predict imminent user a way to opt-out of this tradeoff without feel like you are flying. The VR relies actions and to prefetch the data needed losing the service altogether. The prob- on a smartphone-based HUD, but the for them, today’s systems are less suc- lem is not specific to pervasive comput- Icaros hardware provides the sensor cessful. As mentioned earlier, caching ing; it applies across the broader comput- inputs, which would need to be part of and prefetching run counter to business ing landscape. Consider, for example, a physical smart space. models that are based on forcing the devices such as Alexa and smart toys Also, large wall-sized screens, espe- user to interact with business logic in the that do speech recognition using the cially touch-sensitive screens, are valu- cloud frequently (for example, to present cloud. There is constant risk of leaking able for many kinds of collaboration— advertisements or to track user actions at sensitive information in the use of these the physical scale is a big part of what fine granularity). The obvious solution devices. Good solutions are needed, but makes it valuable, and that is unlikely would be to place a trusted module at I don’t have any silver bullets to offer. to change. the endpoint that performs that busi- A possible solution path is through ness logic locally. This would reduce the the decentralization and local control Have we succeeded in providing “min- frequency of cloud interactions and thus offered by cloudlets. As suggested in imal user distraction”? make the user experience much crisper. the “Privacy Mediators” article I men- We have not been very successful here. Ultimately, the reason to determine user tioned, application-aware third-party If anything, we have worsened the prob- intent is to improve user experience by privacy mediator software on cloud- lem by using business models that rely anticipating user needs and user actions. lets could inspect the raw sensor data primarily on advertising. When writing We still have a long way to go. in real time for potential privacy leaks, the 2001 paper, I was involved in the suppress transmission if a potential leak Aura project, which aimed to reduce user The privacy risks of pervasive com- is detected, and log the instance for a distraction. We did gain some insights puting are significant, as you outlined future privacy audit of the system. This and make some progress, but in hind- many years ago. How much progress is admittedly an imperfect solution and sight, we lacked quantitative measures have we made in the intervening years? involves many moving parts, but it is a to guide our progress. I realize now the Are we building a world in which you start. I think we need to think about wisdom of the aphorism, “That which want to live? this problem in a way similar to deal- gets measured, gets improved.” For a Sadly, I think we are in worse shape ing with viruses on desktops or laptops. systems researcher, it is much easier to regarding privacy than we were in Total elimination of the problem might measure resources such as bandwidth, 2001. The practice of sending all sen- be impossible, but the threat can be energy, and processor utilization than sor data to the cloud makes it too easy reduced to an acceptable level through a fuzzy and ill-defined concept such as for seemingly innocuous data to escape vigilance and preemptive actions. user distraction. So, in 2017, we are no from a user’s control, before he or she better off than we were in 2001 about realizes that sensitive inferences could how to quantify user distraction and be made from that data. The standard Maria R. Ebling is a direc- measure it in a working system without operating procedure today is to ask the tor at the IBM T.J. Watson further distracting the user. This is an user for various privacy-related per- Research Center. Contact important area at the intersection of missions at the time of installing a new her at [email protected]. HCI and systems research. application. This one-time approval is too coarse grain and too far in advance Determining user intent is a significant of necessary context for a user to make challenge. How would you grade our a good decision. Roy Want is a research progress toward this goal? Users need finer grain and more scientist at Google. Con- Our progress has been mixed. Cloud- context-sensitive control of their sensor tact him at roywant@gmail. based machine learning has been very data. As pointed out recently in “Pri- com. successful in, for example, completing vacy Mediators: Helping IoT Cross the user queries to search engines. Being able Chasm,” an article Nigel Davies and I to predict what is being searched for, wrote with Nina Taft, Sarah Clinch, without the user having to type in the and Brandon Amos for HotMobile entire search query, is an example of dis- 2016, cloudlets offer a natural point covering user intent. It is especially valu- of control and enforcement of privacy able in mobile contexts, where a user is policies. Read your subscriptions juggling many things and dealing with a In general, today’s business models through the myCS small screen and a tiny touch keyboard. assume that users are willing to sacrifice publications portal at http://mycs.computer.org. In other areas, such as using location and privacy for free services. There is rarely

JULY–SEPTEMBER 2017 PERVASIVE computing 23