REACHING FOR WEISER’S VISION

Disappearing Hardware

In ’s vision of , computers disappear from conscious thought. From a hardware perspective, the authors examine how far we’ve succeeded in implementing this vision and how far we have to go.

or many tasks today, the use of comput- (such as spell checkers, calculators, electronic trans- ers is not entirely satisfactory. The inter- lators, electronic books, and Web pads). These actions take effort and are often difficult. devices have a specialized interface and address the The traditional, and still prevalent, com- desired goal of ease of use. In contrast, a PC is a gen- puting experience is sitting in front of a eralized machine, which makes it attractive to pur- Fbox, our attention completely absorbed in the dia- chase—the one-time investment having the potential log required to complete the details of a greater task. for many different uses. However, in other ways it Putting this in perspective, the real objective is the adds a level of complexity and formalism that hin- task’s completion, not the interaction with the tools ders the casual user. we use to perform it. Mark Weiser wanted to explore whether we could To illustrate the point further, if somebody asks radically new kinds of computer systems. you for an electric drill, do they want to use a drill, These systems would allow the orchestration of or do they really want a hole? The answer is prob- devices with nontraditional form factors that lend ably the latter—but computers are currently very themselves to more natural, tacit interaction. They much a drill, requiring knowl- would take into account the space in which people 1 and Trevor Pering edge, training, effort, and skill to worked, allowing positional and manipulative — Research, Santa Clara use correctly. Creating a hole is rather than just keyboard and mouse—interactions. relatively simple, and many hid- Along with specialization and the use of embedded Gaetano Borriello den computers invisibly accom- computers, support for mobile computing and wire- University of Washington and plish what seem to be simple less data networks is an important facet of this Intel Research, Seattle tasks, such as regulating our cars’ vision—in other words, invisible connectivity. A goal Keith I. Farkas brakes. But unlike other inani- of this exploration is that we would learn to build Compaq Western Research mate objects, a computer system computer systems that do not distract the user; ide- Laboratory might be able to infer the result ally, the user might even forget the hardware is pres- autonomously and affect the ent.2 In essence, Weiser was proposing that well- desired outcome, increasing both designed computer systems would become invisible its potential and end-user complexity. Realizing this to the user and that our conscious notion of com- potential while managing the complexity is the fun- puter hardware would begin to disappear. Some damental challenge facing computer system years later, popularized this concept researchers. in his book The Invisible Computer.3 An important trend over the last decade is the In this article, we survey the progress toward emergence of specialized, task-specific hardware Weiser’s vision from a hardware viewpoint. Where

36 PERVASIVEcomputing 1536-1268/02/$17.00 © 2002 IEEE Figure 1. Hardware improvement over the last decade: (a) the Xerox ParcTab, the first context-sensitive computer (1992). The design shows the limited display available at that time—a 128- × 64-pixel, monochrome LCD. (b) a typical PDA available today with a color 240- × 320-pixel VGA (transreflective) screen.

have we been, where are we, and where are we headed? What characteristics will make hardware disappear from our conscious- ness, and what will it take to achieve them?

Where we’ve been The research community embraced Weiser’s call to explore ubiquitous com- puting. For example, his vision inspired the work at the Xerox Palo Alto Research (a) (b) Center (PARC) in the early 1990s and such projects as ParcTab,4 Mpad,5 and Liveboard.6 Olivetti Research’s Active a pen and paper. Most of these products use: if we notice it’s there, it’s distracting Badge7 and Berkeley’s InfoPad8 projects have fallen by the wayside: Momenta and us from our real task. For example, if we also embraced this research direction, as EO, IBM’s early ThinkPad, and later the notice that we are using a slow wireless net- did other notable centers of excellence, Apple Newton, the Casio Zoomer, and Gen- work connection instead of just editing our such as at Carnegie Mellon University, eral Magic’s pad. For these , the ben- files, then the action of accessing the files is IBM, Rutgers University, Georgia Tech, efit-to-cost ratio was just not large enough. getting in the way of the real task, which is and the University of Washington. Unfor- To be successful, these new devices had to contained in the files themselves. If the link tunately, many of the early systems were either be better than the traditional pencil- is fast and robust, we will not notice it and based on technologies that were barely and-paper technology they were replacing can focus on the content. Likewise, if a dis- adequate for the task, so they fell short of or provide desired new functionality. The play can present only a poor representa- designer expectations. physical hardware was the dominating fac- tion of a high-quality underlying image, we Figures 1a and 1b illustrate the extent tor, and almost every design aspect affected see a bad display. A high-quality display of hardware improvement over the last acceptance: size, weight, power consump- suspends our belief that the image is only decade. In 1990, no Wireless Local Area tion, computation speed, richness of inter- a representation. Network standards existed; the processors face, and simplicity of design. The four most notable improvements in suitable for mobile devices operated at only We started to cross the acceptability hardware technology during the last decade a few megahertz, while PCs were typically threshold only in the latter half of the that directly affected ubiquitous comput- shipping with up to 50-MHz processors. decade with the Palm Pilot. It was smaller ing are wireless networking, processing The early electronic organizers (pen-based and lighter, focused on simple applications, capability, storage capacity, and high-qual- PDAs had not been invented) proudly and incorporated a novel one-button ity displays. Furthermore, the current pop- claimed 128 Kbytes of memory, while PCs approach to data synchronization. Finally, ular adoption of emerging technology, such shipped with 30-Mbyte disks. The displays an electronic organizer was useful for a sig- as cell phones and PDAs, strongly indicates were also quite crude: laptops used mono- nificant number of people and had real that the market is generally ready for chrome VGA, and the few handheld advantages over the more traditional paper advanced new technology. This adoption, devices available mainly used character- products such as Day-Timers. The com- however, requires common standards based displays. puter industry was beginning to move in across many products and locales. Industry soon responded to the challenge the right direction. with a tighter focus on mobile computing. Wireless networking A flurry of early products hit the market, Where we are Although progress in wireless connec- particularly in the tablet style, that tried to For hardware to disappear from our tivity was initially slow, it has increased. make using computers feel more like using consciousness, we require transparency of This area has witnessed two distinct devel-

JANUARY–MARCH 2002 PERVASIVEcomputing 37 REACHING FOR WEISER’ S VISION

opment trends. The first is in short-range The emergence of the latter IEEE wire- means that we can operate these devices at connectivity standards, such as Bluetooth less standards allows for communication higher speeds, increasing their effective per- (IEEE 802.15) and the IrDA (Infrared Data cells that span many hundreds of feet with formance. Additionally, reduced transistor Association) standards, which are primar- sufficient bandwidth to make us feel as if sizes decrease power consumption, allevi- ily for simple device-to-device communi- we were connected to a wired LAN, but ating some of the perpetual problems sur- cation. Bluetooth, which will get its first without a physical connection’s con- rounding energy storage technologies. real test in the marketplace in 2002, was straints. IEEE 802.11b has already been The combination of more transistors on designed as a short-range cable replace- widely adopted, and 802.11a is expected a given area of silicon and a reduced power ment, allowing for proximate interaction to follow with higher bandwidths. Next- budget has brought us the capabilities of and the discovery of resources in the user’s generation digital cellular networks such mid-1980’s desktop computers in today’s locality. IrDA had a similar aim. But as 2.5G (for example, General Packet battery-operated, handheld PDAs. Two because infrared signaling requires a line Radio Service and NTT’s DoCoMo—with examples are the Motorola Dragonball of sight, users had to physically place greater than 24 million users) and the com- and Intel StrongARM processors, the most devices next to each other, often an incon- ing 3G networks will extend these capa- common processors used by today’s PDAs. venience. This technology, which predates bilities to cover entire metropolitan areas. Besides providing low power consumption Bluetooth by many years, has been con- The wireless networking of today and and high performance, these processors sidered a market failure. (The sidebar lists the immediate future thus enables portable integrate their DRAM and LCD con- URLs for Bluetooth, IrDA, and other areas ubiquitous hardware that remains con- trollers and a host of other interface I/O of interest in this article.) nected to the global infrastructure. How- capability on the same die. These trends directly affect ubiquitous A truly ubiquitous computing experience computing for mobile devices in two ways. First, we can better match algorithmic com- requires high-quality displays to let us see plexity and execution speed to real-world problems. Second, the resulting power con- through the display process and effortlessly sumption allows for a reasonable operating time before batteries fail. These properties acquire the underlying information. have let us build ubiquitous computing hard- ware that we can adapt to a greater range of The second trend is in wireless LAN ever, to date, wireless networks have lagged task-specific activities. Nevertheless, further technology, such as the 11-Mbit-per-sec- behind the bandwidth capabilities of the improvement is still required to satisfy the ond IEEE 802.11b standard and the more equivalent wired networks, leaving both complete ubiquitous computing vision. recent 54-Mbps IEEE 802.11a standard. the opportunity and the user desire for Wireless technologies provide for two improved wireless hardware. Storage capacity basic needs: the ability to detect location A less visible industry trend is the rate and the more basic ability to communicate. Processing capability at which storage capacity is improving for In many cases, the ubiquitous computing For 30 years, processing capability has rotating magnetic storage and solid-state vision can to some degree be implemented basically followed Moore’s law, which can devices. For 25 years, the capacity of by interpreting simple context information, be summarized as “The number of active rotating disks has been roughly doubling such as a user’s location.9 Accurately deter- devices we can place on a given area of sil- every year, a rate of improvement faster mining in-building locations is difficult. icon doubles every 18 months”10 (This is than Moore’s Law! The current storage The Global Positioning System (GPS), for actually a revision of Gordon Moore’s density found on a disk drive is approxi- the most part, can locate an object only to 1965 estimate of doubling per year and the mately 10 Gbits per square inch.11 Today, within 10 meters and does not work in later 1995 estimate of doubling every two IBM markets the Microdrive, a one- buildings. Short-range wireless standards years.) This trend’s obvious consequence square-inch device with 1 Gbyte of stor- such as Bluetooth let us more easily discern is that we can continue to increase the age in a compact flash-card format. If this location and context within a limited area capability of devices fabricated in a given trend continues, in another decade we will without having to support sophisticated area of silicon. Instead of designing systems be able to carry 1 Tbyte of data in a sim- wide-area location technologies. Further- built from separate board-level compo- ilar form factor. more, short range implies lower power and nents, we can integrate diverse functional- From a ubiquitous computing view- the ability to build a smaller device from a ity onto a single chip, resulting in remark- point, storage is becoming so inexpensive more compact energy source, making it ably compact consumer electronics. and plentiful that, for many devices, inter- more attractive for many human-centric Similarly, the reduced capacitance result- nal storage capacity is not a limiting fac- applications. ing from smaller transistor dimensions tor for basic operation. Moreover, we can

38 PERVASIVEcomputing http://computer.org/pervasive Related URLs begin to use storage in extravagant ways Aibo: www.aibo.com by prefetching, caching, and archiving data Blackberry: www.blackberry.net that might be useful later, lessening the Bluetooth: www.bluetooth.com need for continuous network connectivity. DoCoMo: www.nttdocomo.co.jp (in Japanese) Flash memory, DRAM, and Static RAM Electronic ink: www.eink.com have all benefited from progress in inte- IEEE 802.11 Wireless Local Area Networks: gration, making large capacities at low http://grouper.ieee.org/groups/802/11/index.html prices possible. A typical DRAM chip has Infrared Data Association (IrDA): www.irda.org approximately 32 Mbytes of capacity; Intel StrongARM processors: flash memory has 16 Mbytes. Top-end http://developer.intel.com/design/pca/applicationsprocessors/index.htm CompactFlash cards, with 512 Mbytes of Power MEMs research: capacity, closely rival the IBM Microdrive’s http://web.mit.edu/aeroastro/www/labs/GTL/research/micro/micro.html capacity. Universal Plug and Play: www.upnp.org Storage is fundamental to most ubi- Versus Technology: www.versustech.com quitous storage applications. Storage limitations for common tasks hinder the ubiquitous computing experience, forc- ing us to leave behind information and At the smaller end, PDA displays have convergence of previously separate tech- to carefully select what must be mobile. also improved. In the past year, PDAs nologies. Devices such as cell phones, Abundant storage negates these issues, such as the Compaq iPAQ have used trans- PDAs, and digital cameras are beginning letting us focus on the important under- reflective color LCDs. These displays can to merge. Their combined capabilities lying tasks. take natural light entering their surface reduce the number of devices a user must and reflect it back through the LCD stack, carry or own. Such convergent devices will High-quality displays yielding considerable energy savings. likely succeed because numerous devices Because vision is one of our most impor- Although all these displays look promis- will no longer burden users. tant and acute senses, the need to render ing, they still have scope for improvement: In contrast, the problems that a unified information at a very high quality cannot the contrast ratio is lower than most device’s size and complexity create have be overestimated. Low-quality displays printed magazines, and the resolution also given rise to single-function information distract us by drawing our attention to needs to increase to the level of print appliances. These appliances are easily pixelation, granularity, and poor repre- media. adopted because of their simplicity and sentations. Ultimately, the need for improvement in low cost. When personal mobility is not The last decade has seen a remarkable display technology is bounded by the the main motivating factor, such divergent improvement in display technology: most human eye’s visual acuity, but we still have design approaches become attractive—for commercial laptop PCs now have a 13- some way to go. A truly ubiquitous com- example, customizing a computer to be inch color TFT (thin film transistor) LCD puting experience requires high-quality dis- solely a spell checker. Such a device is display at XGA resolution with a viewing plays to let us see through the display physically better suited to its task, but then angle of at least 140 degrees. However, process and effortlessly acquire the under- we have many diverse devices to keep these are mainly transmissive displays lying information. track of and maintain. Even so, organized requiring a backlight that accounts for users can “accessorize” their devices for a approximately one-third of the device’s Adoption trends particular task. There is an ongoing ten- total power consumption. So, from a sys- For the year 2000, the annual sales of sion in ubiquitous-hardware design tem viewpoint, these displays are far from PCs were approximately 150 million between convergent and divergent devices. ideal. Some LCDs have a resolution that units,13 a remarkable statistic about the We can see that our first taste of ubiqui- exceeds 300 dpi, making them suitable for rate of adoption of computer technologies tous computing is already in the PC’s two x-ray-quality pictures. Quality has also in- into all aspects of modern life. However, strongholds—the office and home—and a creased for very large displays, such as the in that year 8 billion embedded processors unique third domain, the automobile. The 60-inch diagonal plasma display that Sam- made their way into the infrastructure of office has benefited from an integration of sung showed at Comdex 2001. Such com- industry and electronic consumer devices. applications and services driven by the mercially available displays provide the So, the fraction of PC sales is a mere 2 per- need for coordinated enterprise solutions. means for shared display workspaces, cent of the total processors sold. From this For example, the Blackberry, a wireless which have been the subject of research statistic, it is obvious that processors are device that integrates with corporate email, for some time (for example, the Stanford beginning to be deployed ubiquitously. has created a considerable following I-Room12). Similarly, another market trend is the among the corporations that have adopted

JANUARY–MARCH 2002 PERVASIVEcomputing 39 REACHING FOR WEISER’ S VISION

it. Such examples show how business pres- where not only the hardware but also the Or, it could even be combined with an sure is pushing the development of inte- results of actions must be removed from the existing device (such as a cell phone). grated systems, which are based on ubi- foreground. Similarly, robotics is an emerg- Extending this model, a user’s personal quitous computing principles. ing field that mobilizes a computer and server could also support interaction The home is also a natural driver for ubi- enables it to effect change at arbitrary loca- through interfaces in the surrounding infra- quitous computing system design. We are tions in the real world. structure. Users could access their personal at an early evolutionary stage where some data through a public kiosk or borrowed homes have established a high-speed net- Personal systems laptop display (see Figure 2b). This work connecting multiple PCs. Consumer Personal systems give users access to extended model is attractive because it products are emerging that exploit this computing independent of their physical allows a favorable (inter- infrastructure—the home computer system location at the cost of them having to carry acting with personal data through a large is slowly subsuming all the home commu- some equipment. Discrete portable devices display) without requiring users to carry nication, entertainment, information deliv- such as PDAs and cell phones are currently the display. The personal server has recently ery, and control systems. the most useful personal systems. How- become tractable owing to advances in The automobile is a particularly out- ever, these systems tend to be limited by short-range wireless technologies (for standing area of success for ubiquitous their computational ability, integration example, Bluetooth) and low-power pro- computing. Modern cars have computer with other devices (for example, your cell cessing (for example, StrongARM), which systems that integrate control of the engine, phone communicating with your PDA), can now support an acceptable user transmission, climate, navigation, enter- and interface capabilities (such as the dis- experience. tainment, and communication systems. play’s size and quality). Today, computa- Success in this domain is largely because tional ability and integration are not hin- Infrastructure systems each car manufacturer has had complete dered by a fundamental limit—advances Unlike mobile systems, infrastructure control over all aspects of its subsystems, in processors and short-range wireless tech- systems instrument a particular locale. So, unlike the home and office domains. Also, nology will eventually solve these prob- many of the difficulties shift from issues of power for computation is not a limitation lems—but interface capabilities are. size, weight, and performance to those of because it is a small fraction of what is Some systems, typically termed wearable deployment, management, and processing. needed to accelerate the car. computers, rely on hardware such as heads- For example, imagine thousands of minia- up displays and one-handed keyboards to ture temperature sensors deployed around Where we’re headed provide the interface to the computer. This a room—How did they get there? How is Ubiquitous computing focuses on getting model is attractive because it provides a the data collected? How are faulty com- computing “beyond the desktop.” This fully functional computing experience ponents identified and replaced?15 Tasks immediately presents a set of research chal- wherever the user might be. However, these that are tractable when the human/com- lenges associated with removing the cus- interfaces can be overly intrusive, requiring puter ratio is close to one suddenly become tomary stationary display-and-keyboard a great deal of the user’s attention; this mit- difficult when the number of computing model, because for most people the PC is igates their widespread acceptance. Cur- devices increases. still their focus. Most current ubiquitous rently, these devices are typically fairly Several hardware projects, such as the computing research projects fall into two bulky belt-worn devices, but they will Berkeley motes,16 have started to explore categories: personal systems, which include shrink as technology progresses, thereby this space by creating a fairly small wireless mobile and wearable systems, and infra- lending themselves to better industrial sensor platform. This lets researchers structure systems, which are associated design and integration. actively explore the networking protocols with a particular physical locale. For both Personal servers14 (see Figure 2a) can en- necessary to organize large sets of nodes. categories, novel interaction modalities, hance ubiquitous access to data. They form Currently about the size of three stacked such as speech or pen processing, become a the computation and storage center of a US quarter-dollar coins, these devices will necessary component because they don’t person’s digital experience. Devices such keep shrinking until they reach the size of require bulky displays or input devices. as a cell phone or PDA-like appliance com- smart dust17 and can no longer be seen or Outside these two categories, the inter- municate directly with this central server, directly manipulated. The Berkeley Pico action of computers with the physical providing a common representation of a Radio project is pursing a single-chip sys- world, without direct human involvement, user’s data. These devices, therefore, merely tem that incorporates both processing and is rapidly becoming more important as the represent an interface into this central radio frequency subsystems.18 total amount of deployable computation repository. With this model, the personal Although the basic hardware of embed- increases. A system’s ability to proactively server could be located out of easy reach— ded devices can be relatively simple, con- monitor and react to the real world is instru- for example, in a user’s shoe, handbag, or siderable hardware challenges remain. mental for truly ubiquitous computing, belt clip—without causing inconvenience. Power is a primary concern. Even though

40 PERVASIVEcomputing http://computer.org/pervasive (a) (b) Personal server

Figure 2. Personal servers: (a) Intel Research’s prototype; (b) using resident display devices such as a wall-mounted display, or a community , to view personal data stored on a user’s personal server.

each node might have a battery lifetime of has no place in systems with either very will greatly increase computers’ impact on several months, the mean time to battery small displays or no display at all. Speech our lives by removing people as a major lim- failure for the entire system can be short if and vision interfaces have done well in iting factor from the processing stream. A it contains many devices. Furthermore, the infrastructure-based systems that have both few automatic systems have already signif- environmental impact of these systems is significant computation resources and a icantly affected our lives: thermostats, air- not well understood—dropping a thou- static environment.19 But they have trou- plane autopilots, automated factories, and sand sensing devices out of an airplane on ble in mobile systems that are computa- antilock brakes, for example. Such systems a disaster zone is relatively easy, but how tionally impoverished or need to operate in still require human intervention: thermostat do you reclaim the devices? dynamic environments that might be, for maintenance, airplane landings, factory con- The fundamental problem with shrink- example, very loud or dark. Complex struction, and deciding when to apply the ing hardware devices is that they quickly touch-based interfaces20 that deal with a brakes. The challenge is to make these sys- become too small and numerous for peo- significant amount of data input or output tems proactive, where they can anticipate ple to relate to them: they are literally out are problematic. They tend to be task spe- and react to physical world conditions (for of sight, so they will quickly become out of cific, with interface hardware crafted for a example, deciding to apply the brakes), mind. One direct byproduct of these mul- specific application, so they have not gained instead of just reacting to them (for exam- tidevice systems is that the individual net- wide acceptance. ple, deploying an air bag when you crash).21 work nodes will not be named in any The most exciting advances in ubiqui- The salient distinction between these two human-understandable way—there will tous interfaces will likely be new display models is that one is human-centric, which simply be too many of them to keep track technologies that enable rich visual output requires close involvement to effect correct of. System security rapidly becomes a big without a bulky flat-screen display. E-Ink operation, while the other is human-super- concern: how do you know if a node should (electronic ink) Corp., for example, uses a vised, requiring minimum involvement but be listening in on a wireless conversation if system of microcapsules to create flexible still achieving an intelligent, useful result. you can’t even keep track of which nodes display surfaces, which would be consider- A proactive system must closely and reli- you have in the system? ably more convenient than a traditional ably integrate sensors and actuators with rigid LCD panel. Such technology, in con- the physical world. This task is closely Proactive interaction junction with an abstracted computation related to building the infrastructure-based Both personal and infrastructure-based model such as the personal server, brings us systems we described earlier. However, systems ultimately require some kind of one step closer to a world where we can proactive systems will require greater user interface to let humans interact with access personally relevant information sophistication in the components deployed them. Nondesktop interface modalities, quickly and conveniently, without relying in the environment, both to enable the such as pen, speech, vision, and touch, are on bulky, fragile display systems. Interface capability to affect the physical world and attractive in ubiquitous computing systems hardware technology poses different diffi- to quickly, robustly, and accurately pro- because they require less of a user’s atten- culties for mobile systems than for desktop cess real-world data. For example, for a tion than a traditional desktop interface. computers. For example, speech interfaces building-scale temperature-monitoring However, the mainstream use of these tech- for mobile devices are inherently problem- application, slowly reporting distributed niques depends largely on their hardware atic because they often operate in noisy temperatures to a central server would be requirements and interface capabilities. environments, requiring noise cancellation sufficient. However, an earthquake re- For example, pen computing has largely or directional-microphone techniques. sponse system would need to actively been successful on PDA-class devices with Directly integrating computing with the dampen vibrations at many nodes through- a well-defined and accessible display. But it real world, without human intervention, out a building.

JANUARY–MARCH 2002 PERVASIVEcomputing 41 REACHING FOR WEISER’ S VISION

location problem mentioned in the previ- ous paragraph.

Hard problems & physical limitations Three hard problems have been the core challenge of ubiquitous computing hard- ware and will likely continue to be far into the future: size and weight, energy, and the user interface. These problems transcend individual points on the technology curve, partly because they are somewhat contra- dictory—as we’ll see, a solution in one space greatly confounds that in another. Ongoing solutions to these problems, therefore, will have to come from “outside the box” and remove the fundamental roadblocks.

Size and weight Size and weight significantly impede Figure 3. Sony’s Aibo robot dog. ubiquitous computing because they con- tinually remind the user of the hardware’s presence. This limitation manifests itself From a personal perspective, proactive namely, Sony’s Aibo (see Figure 3)—have both in mobile systems when the user must computing pushes us toward systems that reached the market. Abstractly, robots are carry the device and during the setup and can monitor and affect our bodies directly. a novel type of disappearing hardware: they configuration of infrastructure-based sys- For example, automatically handling dia- allow computation to directly affect the real tems involving many pieces of equipment. betes requires the ability to both monitor world without heavily instrumenting the Appropriately, the two main contributors blood glucose levels and administer environment. to a device’s size and weight stem from the insulin—tasks requiring specialized hard- Software issues aside, significant hard- other two fundamental problems: batteries ware. A system such as this would still ware challenges exist, particularly for the and the user interface. require significant advances in software sensing technologies that will enable proac- The Itsy pocket computer (see Figure reliability to be feasible. tive robotics. Vision is the canonical tech- 4)22 exemplifies this situation. It is just 70 nique for a robot to determine where it is, percent larger than its 2.2 watt-hour bat- Robotics what objects are around it, and where they tery and 320- × 200-pixel display. In addi- Now more popular than ever, robotics are. The generalized vision problem re- tion, these two components represent 60 presents an interesting confluence between quired to deal with dynamically changing percent of the total weight without the case mobile and proactive systems: allowing a physical locations is quite difficult. One but 43 percent with the case. computer system to affect the real world solution is to instrument a particular envi- Overall, device sizes have been shrink- without a priori instrumentation of the ronment with sensors, beacons, or both to ing at an incredible rate owing to system- environment. Most robotic systems either aid a robot in a particular locale—the level integration. But devices are reaching perform a very specific task, such as factory equivalent of GPS for buildings. A high- the point where they can’t get much smaller automation, or are remotely controlled by accuracy indoor-positioning system would and still be usable or provide additional a person. In general, autonomous robots better allow an autonomous robot to func- benefit. Moreover, such reductions might have difficulties dealing with dynamic phys- tion within a building’s confines. Addi- increase, rather than decrease, the cogni- ical situations—primarily with determin- tionally, tagging interesting objects in the tive load. For example, many new cars ing where they are, as well as with identi- environment would help a robot identify include a key fob for locking the doors and fying objects in the environment. Similar to and locate them. By exploiting infrastruc- arming the alarm. If the fob were the size handheld devices, the hardware for con- ture-based computing, as we discussed ear- of a brick, few people would use it because structing robots is shrinking rapidly, mak- lier, robotics can make significant strides it would be too large. Alternatively, if it ing them cheaper and more capable. toward being truly proactive and auto- were the size of a penny, few people would Recently, the first real consumer robots— nomous without solving the generalized use it owing to difficulty manipulating the

42 PERVASIVEcomputing http://computer.org/pervasive Figure 4. The Itsy pocket computer exemplifies the trend that the size and weight of energy sources and I/O devices is dominating the size and weight of mobile computers.

buttons. In this case, the most suitable size and weight are inextricably tied to the device’s intended function.

Energy The challenge for any mobile system is to reduce user involvement in managing even generating, energy. For example, MIT and cannot autonomously signal their pres- its power consumption. Energy is a nec- researchers have exploited the human body ence. The ability to transmit energy, when essary resource for virtually all comput- as an energy source by constructing sneak- combined with robotics, gives us the capa- ing systems, but any reference to it ers that use flexible piezoelectric structures bility for mobile computation that can detracts from a positive user experience. to generate energy (see Figure 5).26,27 Sim- recharge itself. The degree to which an energy source dis- ilarly, solar radiation, thermal gradients, tracts the user depends on the intended mechanical vibration, and even gravita- User interfaces application, the hardware implementing tional fields all represent potential power Rendering user interfaces invisible is fun- the application, and the energy source’s sources for a .15 damentally difficult owing to the tradeoff characteristics. Additionally, storage technologies under between size and weight and . Solutions to this problem fall into two development promise much greater energy Reducing the size and weight will make the approaches. The first is to reduce power con- densities than those of conventional bat- device less visible but might decrease its sumption. Part of this approach consists of teries. In the near term, one of the more usability. The degree to which these two designing energy-aware software that can promising technologies is fuel cells, partic- components matter depends on the specific identify the hardware states that provide a ularly direct methanol fuel cells.28 Pure properties of the interface and of the appli- given service level and select those that are methanol fuel offers an energy density cation for which it is being used. most energy efficient. For instance, in sys- roughly 40 times that of a Lithium-ion For example, to open or lock a door, a tems with a microprocessor whose energy polymer battery, but 70 to 90 percent of user might use a remote control contain- consumption is greater at high speeds, the this chemical energy is lost in conversion to ing one button that unlocks and opens the software can select the lowest speed possi- electricity. In the longer term, technologies door when pressed once and locks it when ble that still achieves the required task’s per- such as MIT’s MEMS (microelectro- pressed twice in quick succession. Because formance.23 The control software can also mechanical systems)-based microturbine the single button serves multiple purposes, modify the quality of service it seeks to and associated micro electric generator users will likely find it more demanding deliver.24 For instance, to save energy, the might provide highly compact energy than the original interfaces (the door knob software could reduce the frame rate, or size, sources with significantly longer lifetimes. and door lock). Alternatively, the remote of an MPEG movie, incrementally resulting An alternative to acquiring energy is to control could have two buttons, one for in a corresponding loss of fidelity. Or, the transmit energy to a mobile device, reduc- opening and one for locking and unlocking software could forward a voice utterance to ing its need for an autonomous power the door. At the expense of increased size a remote system for recognition rather than source. This technique is difficult to do and weight, this solution substitutes a spa- expending local energy on the task.24 Soft- safely over a long range, but it is applica- tial differentiation (two single-function ware systems are just beginning to address ble to the field of passive electronic tagging. buttons) for a temporal one (a multiple- these issues concerning energy awareness. For example, Radio Frequency Identifica- function button). These two solutions rep- The second approach is to find alterna- tion29 tags are inductively powered by the resent the chief user-interface tradeoff that tive and improved energy sources. tag reader, typically up to a maximum of makes good user interface design for small Although battery energy densities are pro- one meter, employing load modulation to ubiquitous devices fundamentally hard. jected to increase approximately 10 per- transmit their data back to the interroga- Today’s most popular user interfaces cent annually for the next three years,25 the tor. Such passive tags have unlimited life- include buttons, keyboards, mice, point- energy storage costs of batteries will likely times, are smaller, and cost less; however, ers, LCD panels, touch screens, micro- remain significant. This will lead us to unlike battery-powered (active) tags, they phones, and speakers. These elements are explore other technologies for storing, and can communicate only over a short range designed for high-rate information flow,

JANUARY–MARCH 2002 PERVASIVEcomputing 43 REACHING FOR WEISER’ S VISION

each suited to a specific application class. where the computer and the real world are devices that use the baseline minimum For example, a keyboard and display seem tightly integrated. Nonetheless, direct power for the job or occupy the smallest ideal for writing a book, but a microphone neural interfaces embody tremendous risks, possible volume. Processors will continue and speaker would probably be better for not the least of which is loss of human to shrink while increasing in capability and communicating with another person. For autonomy. Clearly, this area requires much capacity. However, new applications will other applications, such as alerting a user more research, but if we can overcome the demand ever-greater processing capabilities, that he or she has received mail, these user challenges, such interfaces would go a long not the least of which will be those of com- interfaces are unnecessarily complex. way toward reaching Weiser’s vision. munication and security. For example, Other, more unobtrusive, mechanisms such implementing public key encryption with a as the ambientROOM,30 explore how to Future challenges typical microcontroller instruction set leads communicate low-latency, low-importance We will soon be able to include com- to large code size, high power requirements, information via interfaces that require lit- puter hardware into virtually every man- and slow performance.33 The challenge will tle direct attention. For example, variances ufactured product, and provide a wireless be to include appropriate primitives that in a projected image’s intricacy could indi- infrastructure to let these devices commu- make these operations more efficient in all cate if there is unread mail, thereby mak- nicate directly or indirectly. But what will these dimensions while permitting the evo- ing information available to the user unob- they communicate about and in what pro- lution of security algorithms. trusively through a directed glance. tocol? How will a user or user’s applica- As we crowd more and more frequencies Although these interfaces are distinctly tion know what devices to use for what with wireless communications, we will need separate from our bodies, the natural direc- purpose? How will a user know when to make our radio systems adaptive. Devices tion for disappearing interfaces is for the invisible devices are present, functioning, will need to adjust their bandwidth require- two to blend together. Several researchers and not compromising privacy? What ments on the basis of what other devices are are exploring the possibility of interpret- actions can a user take if the situation is in the radio neighborhood. Such a necessity highlights the critical problem of system evo- A key element of ubiquitous computing lution. How will the development and deployment of new devices affect the devices applications is knowing the precise we already have deployed? Flexibility will be needed, not only in terms of software spatial–temporal relationships between updates to adjust how a device is used but also in terms of wireless communication. people and objects. Software radios are promising in this sec- ond dimension; they let a device change how ing information from our neurons to let us unacceptable? These questions suggest that it uses the spectrum to be compatible with control computers and other machines by hardware, software, user interaction, and its neighbors.34 just thinking about doing so.31 Early work applications all have unresolved issues that involving a monkey with neural implants we must address before ubiquitous com- Disappearing software has demonstrated the ability to gather suf- puting will truly reach Weiser’s goal of To make hardware disappear, we also ficient information to let a robot mimic the improving, rather than further complicat- need to make software disappear. Today’s monkey’s arm movement.31 Similarly, ing, our lives. software is too monolithic and stovepiped, neural implants can feed information into and is written with many assumptions the brain, removing the need for humans to Unresolved hardware issues about hardware and software resources. gather information through their senses. We have already discussed some of the The ubiquitous computing environment This approach’s potential is suggested by challenges for our hardware platforms. that we must create will be much more recent research employing cochlear Specifically, we will need to continue to dynamic. Devices, objects, and people will implants to help those with hearing loss to manage our devices’ power requirements. be constantly moving around, creating an communicate better and become more Making devices ubiquitous can’t be coupled ever-changing set of resources—different aware of their surroundings.32 For exam- with the need to change batteries, or even user input/output interfaces, displays, and ple, such neural implants could bring infor- recharge them, when thousands of devices windowing systems—that will be available mation (such as “you’ve got mail”) to the are involved. The solutions presented ear- to our applications. Also, some devices user’s attention by fooling the brain into lier only partially solve the problems; some might lose their ability to communicate thinking a subtle but noticeable image are highly experimental and might not be with others owing to interference or envi- has been projected onto the retina. practical for unforeseen reasons. ronmental conditions. How will we con- These interfaces form the personal-inter- Microprocessors can also continue evolv- struct applications to operate in such envi- action version of proactive computing, ing. The need will persist for minimal ronments?

44 PERVASIVEcomputing http://computer.org/pervasive Figure 5. Energy-scavenging shoes using piezoelectric material, developed at MIT’s Media Laboratory. With these shoes, normal walking motion can generate sufficient energy to broadcast an ID every three to five steps.27 Photo courtesy of the MIT Media Laboratory.

To do this, we must even further decou- ple our applications into small pieces of code, spread across ubiquitous hardware, that can come together as needed and expect to have connections constantly enabled and disabled. We must create data interchange formats where the data not software handle multiple users, but also structure,40,41 or are based on proximity only is self-describing but also can find its the hardware must be able to accurately (for example, electronic tags). We desper- way from the device that created it to its identify and differentiate between the mul- ately need tags that can be located within destination as autonomously and securely tiple users. Addressing these issues using a few millimeters, are cheap to create (for as possible.35 We must develop mecha- what we’ve learned from the desktop example, by printing), and are completely nisms for devices to advertise their capa- metaphor seems less promising because passive. Ideally, we would also want the bilities so that applications, whether in the that interface has not been developed or ability to detect that two tagged objects are infrastructure or on portable devices, can optimized for casual use by multiple users. physically touching rather than just in close become aware of and select from the With computing capability in every proximity.42,43 Coming up with the tech- panorama of available resources. object, users will want to take advantage nologies that provide these capabilities, Discovery services36,37 (for example, Uni- of the devices they encounter throughout even if initially imperfectly, is another key versal Plug and Play) to some extent already their day without worrying about owner- challenge. For more on this topic, see enable such cyber foraging.38 These first- ship or security at every step. Consider how “Connecting the Physical World with Per- generation systems require highly capable we use paper and pencil: we can easily jot vasive Networks,” in this issue. devices that can download code and enjoy down notes and later identify the author stable and relatively high-bandwidth con- based solely on the handwriting. The note- Applications nections. This approach will not scale to taking process directly incorporates this Currently, most applications are based the myriad sensing and processing devices process; there is no explicit authentication on ownership of relatively large, multi- that will surround us. They will employ mechanism. Similarly, physical control purpose hardware devices with only the minimal computational elements and exist guarantees privacy: we put the paper in our most limited interaction with the physical in small communication cells with highly pocket and can hide it from others’ eyes. world in which people live. We need to dis- variable communication and processing What metaphors will we have for the elec- cover and enable those most compelling properties. This scenario requires that mul- tronic paper that can communicate with applications that will let us deploy the first tiple devices replace a single device. Adap- other devices? Without some kind of phys- truly ubiquitous systems. Yet, as we have tation at this level is another challenge we ical icon, how do we seamlessly control already discussed, the most difficult part must tackle to achieve long-lived systems sharing content with other people? of this will be to give these systems an evo- that can evolve gracefully. A key element of ubiquitous computing lutionary path, in contrast to today’s applications is knowing the precise spatial– approach of completely reengineering all temporal relationships between people and the component devices. Software engi- Although new technologies are creating objects. Such knowledge succinctly helps neering will take center stage in this effort. new ways of interacting with our compu- specify intent, an integral component of Clearly, we need a new set of abstractions tations, we could end up trading the prob- user interfaces. But the resolution of loca- to make writing such applications possi- lems of one user interface for those of tion systems needs to improve dramati- ble—abstractions that support software another. For example, wireless technolo- cally.39 Current commercial indoor loca- deployment in separate modules on thou- gies permit a device we are carrying to tion systems are either too coarse, sands of devices, cyber foraging, and hard- interact with a public display. But what operating primarily at the room level7 (for ware sharing. This is a different world of happens when multiple potential users are example, the Versus Information System), software development than we are accus- surrounding a display? Not only must the or require prohibitively expensive infra- tomed to.

JANUARY–MARCH 2002 PERVASIVEcomputing 45 REACHING FOR WEISER’ S VISION

any hardware components on the hard problems associated with ubiquitous Trends through 2004, MPUs, MCUs, DSPs necessary to build ubiqui- computing. And a special thank you to Mark Weiser, and Cores, Gartner Group, Stamford, in memory, for his visionary ideas in the early 1990s. Conn., 22 Jan. 2001. tous computing systems are now available. Key 14.R. Want and B. Schilit, “Guest Editors’ improvementsM since Weiser’s original vision Introduction: Expanding the Horizons of REFERENCES Location-Aware Computing,” Computer, of ubiquitous computing include wireless vol. 34, no. 8, Aug. 2001, pp. 31–34. networks; high-performance and low- 1. K. Fishkin et al., “Embodied User Interfaces power processors; high-quality displays; for Really Direct Manipulation,” Comm. 15. A. Chandrakasan et al., “Design Consider- ACM, vol. 43, no. 9, Sept. 2000, pp. 75–80. ations for Distributed Microsensor Sys- and high-capacity, low-power storage tems,” Proc. 1999 IEEE Custom Integrated devices. The progress toward building soft- 2. M. Weiser and J. Seely-Brown, “The Com- Circuits Conf., IEEE Press, Piscataway, N.J., ware systems to orchestrate these compo- ing Age of Calm Technology,” Beyond Cal- 1999, pp. 279–286. culation: The Next Fifty Years of Comput- nents has been less dramatic, with many ing, P.J. Denning and R.M. Metcalfe, eds., 16. J. Hill et al., System Architecture Directions unresolved issues relating to user interfaces, Copernicus, Heidelberg, Germany, 1998. for Network Sensors, Proc. 9th Int’l Conf. security, privacy, and managing complex- Architectural Support for Programming 3. D.A. Norman, The Invisible Computer, Languages and Operating Systems (ASPLOS ity. Consequently, the hardware compo- MIT Press, , Mass., 1998. 2000), ACM Press, New York, 2000, pp. nents for many applications are reaching 93–104. the point where a user is less likely to be 4. R. Want et al., “An Overview of the ParcTab Ubiquitous Computing Experiment,” IEEE 17.J.M. Kahn, R.H. Katz, and K.S.J. Pister, distracted by the medium than by the inter- Personal Comm., vol. 2, no. 6, Dec. 1995, “Next Century Challenges: Mobile Net- action with the controlling software. And pp. 28–43. working for ‘Smart Dust,’” Proc. ACM/ as a result, at this level, the hardware is IEEE Int’l Conf. Mobile Computing and 5. C. Kantarjiev et al., “Experiences with X in Networking (MobiCom 99), ACM Press, beginning to become invisible. a Wireless Environment,” Proc. Usenix New York, 1999, pp. 271–278. Of course, hardware could become too Symp. Mobile & Location-Independent invisible. At some point, you might need Computing, Usenix Assoc., Berkeley, Calif., 18. J.M. Rabaey, “Wireless beyond the Third 1993, pp. 117–128. to know what is real and what is simulated, Generation—Facing the Energy Challenge,” Proc. 2001 Int’l Symp. Low Power Elec- and have some handle on an interface’s 6. S. Elrod et al., “Liveboard: A Large Interac- tronics and Design (ISLPED 01), ACM location and components. However, tech- tive Display Supporting Group Meetings, Press, New York, 2001, pp. 1–3. Presentations, and Collaborations,” Proc. nological advances clearly will continue to 1992 Conf. Human Factors in Computing 19. J. Crowley, J. Coutaz, and F. Berard, “Per- bring us new hardware components that, Systems (CHI 92), ACM Press, New York, ceptual User Interfaces: Things That See,” if we wish, will function more invisibly 1992, pp. 599–607. Comm. ACM, vol. 43, no. 3, Mar. 2000, pp. 54–64. than before, until every aspect of an inter- 7. R. Want et al., “The Active Badge Location face crosses a threshold that no longer hin- System,” ACM Trans. Information Systems, 20.H. Ishii and B. Ullmer, “Tangible Bits: ders our senses. We will begin to see vol. 10, no. 1, Jan. 1992, pp. 91–102. Towards Seamless Interfaces between Peo- through it, just as we see through the ink of ple, Bits and Atoms,” Proc. 1997 Conf. 8. T. Truman et al., “The InfoPad Multimedia Human Factors in Computing Systems (CHI printed text and focus on the information Terminal: A Portable Device for Wireless 97), ACM Press, New York, 1997, pp. it contains. At that point, by definition, the Information Access,” IEEE Trans. Com- 234–241. puters, vol. 47, no. 10, Oct. 1998, pp. hardware will have disappeared. When 1073–1087. 21. D. Tennenhouse, “Proactive Computing,” that day comes, computer hardware will Comm. ACM, vol. 43, no. 5, May 2000, p. likely be mediating in every aspect of our 9. B. Schilit, N. Adams, and R. Want, “Con- 43. text-Aware Computing Applications,” Proc. daily activities, and Weiser’s vision will be Workshop Mobile Computer Systems and 22. W.R. Hamburgen et al., “Itsy: Stretching the almost complete. Applications, IEEE CS Press, Los Alamitos, Bounds of Mobile Computing,” Computer, Calif., 1994, pp. 85–90. vol. 34, no. 4, Apr. 2001, pp. 28–36.

10. G. Moore, “Cramming More Components 23. M. Weiser et al., “Scheduling for Reduced ACKNOWLEDGMENTS onto Integrated Circuits,” Electronics, vol. CPU Energy,” Proc. 1st Symp. Operating 38, no. 8, 19 Apr. 1965, pp. 114–117. Systems Design and Implementation, The thoughts and views expressed in this article Usenix, Berkeley, Calif., 1994, pp. 13–23. were not formed in isolation. We thank the many 11. J.W. Toigo, “Avoiding a Data Crunch,” Sci- colleagues who have worked alongside us on ubi- entific American, vol. 282, no. 5, May 2000, 24. J. Flinn, Extending Mobile Computer Bat- quitous computing projects and, in doing so, have pp. 58–74. tery Life through Energy-Aware Adaptation, shaped our views in this exciting research area. PhD dissertation, tech. report CMU-CS-01- Particularly, we thank Sunny Consolvo, Gunner Dan- 12.A. Fox et al., “Integrating Information 171, Dept., Carnegie Mel- neels, and the IEEE Pervasive Computing editorial Appliances into an Interactive Work- lon Univ., Pittsburgh, 2001. board and supporting reviewers for their comments; space,” IEEE Computer Graphics & Ap- Jim Kardach, Graham Kirby, Muthu Kumar, Murali plications, vol. 20, no. 3, May/June 2000, 25. Y. Kim, “Capacity Enhancement Technol- Sundar, and Paul Wright for the presentation of the pp. 54–65. ogy of Samsung Soft Case Li-ion Batteries,” personal server prototype; and Bob Mayo and Bill Proc. Power 2001: 9th Ann. Int’l Conf. Hamburgen for their contributions to our discussion 13. Embedded Microcomponent Forecast and Power Requirements for Mobile Comput-

46 PERVASIVEcomputing http://computer.org/pervasive ing, Wireless Electronic Devices & Busi- puter, vol. 34, no. 8, Aug. 2001, pp. 57–66. and Technology (UIST 2001), ACM Press, ness/Consumer Applications, 2001. New York, 2001, pp. 183–190. 40. M. Addlesee et al., “Implementing a Sentient 26.T. Starner, “Human-Powered Wearable Computing System,” Computer, vol. 34, no. 43. B. Vigoda and N. Gershenfeld, “TouchTags: Computing,” IBM Systems J., vol. 35, nos. 8, Aug. 2001, pp. 50–56. Using Touch to Retrieve Information Stored 3–4, 1996, pp. 618–629. in a Physical Object,” Extended Abstracts 41. A. Ward, A. Jones, and A. Hopper, “A New of Computer-Human Interaction 1999, 27. N. Shenck and J. Paradiso, “Energy Scav- Location Technique for the Active Office,” ACM Press, New York, 1999, pp. 264–265; enging with Shoe-Mounted Piezoelectrics,” IEEE Personal Comm., vol. 4, no. 5, Oct. www.media.mit.edu/physics/publications/ IEEE Micro, vol. 21, no. 3, May/June 2001, 1997, pp. 42–47. papers/99.01.touchtag.rtf. pp. 30–42. 42. K. Partridge et al., “Fast Intrabody Signal- 28. S. Thomas and M. Zalbowitz, Fuel Cells— ing: Empirical Measurements of Intrabody Green Power, Los Alamos Nat’l Lab, Los Communication Performance under Var- For more information on this or any other com- Alamos, N.M., 1999; www.lanl.gov/energy/ ied Physical Configurations,” Proc. 14th puting topic, please visit our Digital Library at est/transportation/trans/pdfs/fuelcells/fc.pdf. Ann. ACM Symp. User Interface Software http://computer.org/publications/dlib.

29. K. Finkenzeller, The Radio Frequency Iden- tification (RFID) Handbook, John Wiley & Sons, New York, 1999.

30. H. Ishii et. al., “AmbientROOM: Integrat- ing Ambient Media with Architectural Space,” Proc. 1998 Conf. Human Factors in Computing Systems (CHI 98), ACM Press, New York, 1998, pp. 173–174. the AUTHORS 31.A. Regalado, “Miguel Nicolelis: Brain- Machine Interfaces,” Technology Rev., Roy Want is a principal engineer at Intel Research. His interests include ubiquitous Jan./Feb. 2001, pp. 98–100. computing wireless protocols, hardware design, embedded systems, distributed systems, and automatic identification and microelectromechanical systems. While at 32.M. Nicolelis, “Actions from Thoughts,” Olivetti Research, he developed the Active Badge, a system for automatically locating Nature, vol. 409, no. 6818, 18 Jan. 2001, people in a building. As part of Xerox PARC’s Ubiquitous Computing program, he led pp. 403–407. the ParcTab project, one of the first context-aware computer systems. At PARC, Want also managed the Embedded Systems group. He received his BA and PhD in computer 33. A. Perrig, “SPINS: Security Protocols for science from Churchill College, Cambridge University. Contact him at Intel Corp., 2200 Sensor Networks,” Proc. 7th Ann. ACM Mission College Blvd., Santa Clara, CA 95052; [email protected]; www.ubicomp.com/want. Int’l Conf. Mobile Computing and Net- working (MobiCom 2001), ACM Press, Trevor Pering is a research scientist at Intel Research. His research interests include New York, 2001, pp. 189–199. many aspects of mobile and ubiquitous computing, including usage models, power management, novel form factors, and software infrastructure. He received his PhD 34. V. Bose, D. Wetherall, and J. Guttag, “Next in from the University of , Berkeley, with a focus on Century Challenges: RadioActive Net- operating system power management. He is a member of the ACM. Contact him at works,” Proc. ACM Mobile Computing and [email protected]. Networking (Mobicom), ACM Press, New York, 1999, pp. 242–248.

35. M. Esler et al., “Next Century Challenges: Gaetano Borriello is a faculty member of the University of Washington’s Department Data-Centric Networking for Invisible Com- of Computer Science and Engineering. He is on a two-year leave to establish a new puting—The Portolano Project at the Uni- Intel research center adjacent to the University of Washington campus. His research versity of Washington,” Proc. MobiCom interests are in the design, development, and deployment of computing systems with 1999, ACM Press, New York, 1999, pp. particular emphasis on mobile and ubiquitous devices and their application. His most 256–262. recent research accomplishments include the development of the Chinook design system for heterogeneous distributed embedded processors. He received his BS 36.W. Adjie-Winoto et al., “The Design and in electrical engineering from the Polytechnic Institute of New York, his MS in elec- Implementation of an Intentional Naming trical engineering from , and his PhD in computer science from the University of System,” Operating Systems Rev., vol. 34, California, Berkeley. He received an NSF Presidential Young Investigator Award in 1998 and a University no. 5, Dec. 1999, pp. 186–201. of Washington Distinguished Teaching Award in 1995. Contact him at the Dept. of Computer Science & Eng., Univ. of Washington, Box 352350, Seattle, WA 98195-2350; [email protected]. 37. J. Waldo, Jini Architecture Overview, tech. report, Sun Microsystems, Palo Alto, Calif., Keith I. Farkas is a senior member of the research staff of Compaq Computer’s Jan. 1999. Western Research Lab. His research interests include microprocessor design and software and hardware techniques for managing and optimizing computer system 38. M. Satyanarayanan, “Pervasive Computing: energy consumption. He received his PhD from the University of Toronto. He is a Vision and Challenges,” IEEE Personal member of the IEEE and ACM. Contact him at [email protected]. Comm., vol. 8, no. 4, Aug. 2001, pp. 10–17.

39. J. Hightower and G. Borriello, “Location Systems for Ubiquitous Computing,” Com-

JANUARY–MARCH 2002 PERVASIVEcomputing 47