Context-Aware Power Management

Abstract

With more and more computing devices being deployed in buildings therehas been a steady rise inbuildings’ electricity consumption. These devices not only consumeelectricity but also produce heat,which increases loading on ventilation systems, further increasingelectricity consumption. At thesame time there is a pressing need to reduce overall building energyconsumption. For example, theEuropean Union’s strategy for security of energy supply highlightsenergy saving in buildings as akey target area. One approach to reducing energy consumption of devicesin buildings is to improvethe effectiveness of their power management.Current state-of-the-art computer power management is predominantlyfocused on extending bat-tery life for mobile computing devices. The majority of policies arelow-level and are used to managesub-components within the overall computing device. The key trade-offfor these policies is deviceperformance versus increased battery life. In contrast, stationarycomputing devices do not have bat-tery limitations and typically the most significant energy savingsare achieved by switching the entiredevice to standby. However, switching to a deep standby state cancause significant user annoyancedue to the relatively long resume time and possible false power downs.Consequently these energysaving features are typically not enabled (or used with long timeouts).To increase enablement, poli-cies for stationary devices need to operate in a near transparentfashion, i.e., operate automaticallyand with little user-perceived performance degradation.Context-aware pervasive computing describes a vision of computingeverywhere that seamlesslyassists us in our daily tasks, i.e., many functions are intelligentlyautomated. Information display,computing, sensing and communication will be embedded in everydayobjects and within the environ-ment’s infrastructure. Seamless interaction with these devices willenable a person to focus on theirtask at hand while the devices themselves vanish into the background.Realisation of this vision couldexacerbate the building energy problem as more stationary computingdevices are deployed but itcould also provide a solution. Context information (e.g., user locationinformation) likely to be avail-able in such pervasive computing environments could enable highlyeffective power management formany of a building’s electricity consuming devices. We term such powermanagement techniques ascontext-aware power management (CAPM), their principal objective beingto minimise overall elec-tricity consumption while maintaining user-perceived device performance.The current state of theart in context-aware computing focuses on developing inference techniquesfor determining high-levelcontext from low-level, noisy, and incomplete sensor data. Possibleapproaches include rule-basedinference, Bayesian inference, fuzzy control, and hidden Markov models.Successful inference enablesthe vision of computing services interfacing seamlessly and transparentlywith users’ daily tasks. Onesuch desirable, transparent service is context-aware power management.We have identified several key requirements and designed a frameworkfor CAPM. At the coreof the framework, a Bayesian inference technique is employed to inferrelevant context from a givenrange of sensors. We have identified the principal context requiredfor effective CAPM as being (i)when the user is not using and (ii) when the user is about to usea device. Accurately inferringthis user context is the most challenging part of CAPM. However, thereis also a balance betweenhow much energy additional context can save and how much it will costboth monetarily and energywise. To date there has been some research in the area of CAPM butto our knowledge there hasbeen no detailed study as to what granularity of context is appropriateand what are the potentialenergy savings.We have conducted an extensive user study to empirically answer thesequestions for CAPM ofdesktop PCs in an office environment. The sensors used are keyboard/mouseinput, user presencebased on Bluetooth beaconing, near presence based on ultrasonic rangedetection, face detection, andvoice detection. Results from the study show that there is wide variabilityof usage patterns andthat there is a balance whereby adding more sensors actually increasesthe energy consumption. Forthe desktop PC study, idle time, user presence, and near presenceare sufficient for effective powermanagement coming within 6-9% of the theoretical optimal policy (onaverage). Beyond this facedetection and voice detection consumed more than they saved. The evaluationfurther demonstratesthe use of Bayesian inference as a viable technique for CAP