1 / 32

Context-aware Battery Management for Mobile Phones ( PerCom 08)

Context-aware Battery Management for Mobile Phones ( PerCom 08). Nishkam Ravi, James Scott, Lu Han and Liviu Iftode. 이상훈 , 오교중 2009. 12. 07. Contents. Introduction Problem definition System design Evaluation Conclusion Pros. Introduction.

kay
Télécharger la présentation

Context-aware Battery Management for Mobile Phones ( PerCom 08)

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Context-aware Battery Management for Mobile Phones (PerCom 08) Nishkam Ravi, James Scott, Lu Han and LiviuIftode 이상훈, 오교중 2009. 12. 07

  2. Contents • Introduction • Problem definition • System design • Evaluation • Conclusion • Pros

  3. Introduction • Mobile devices are providing increasing functionality due to rapid improvements • However, battery capacities are not improved as other technologies • Energy will remain the main bottleneck in the future

  4. Introduction • Current battery management • Informed to decide prioritization of the tasks • Battery meter • “battery low” audio signals • Remaining time estimate at current power • The user get into habit of charging at suitable period • Based on their call patterns • Low-power standby modes • Accustomed to the users

  5. Introduction • Factors to change current battery management • Multi-functional computing expects always-on • WLAN are hungry consumer of energy • Pervasive computing asks to be always-on for background applications • These battery consumptions • Require the user to charge more frequently • Break the low standby-mode power profile

  6. Introduction • Goal • Propose a new context-aware battery management architecture for mobile devices (CABMAN) • Three principles • Crucial applications (telephony) should not be compromised by non-crucial applications • Charging opportunities should be predicted • Context can be used to predict charging opportunities

  7. Problem Definition • Will the phone battery last until the next charging opportunity is encountered? • When the next opportunity for recharging the battery will be available? • If then what is the total battery lifetime available to the user? • What fraction of this battery lifetime will be consumed by critical applications such as telephony? • What fraction of this battery lifetime can be left for use by noncritical applications?

  8. Problem Definition • To build a system that can monitor user context and sense the battery charge level of the device, it requires • Aset of algorithms for making predictions • Acentral component for assimilating the information together and warning the user appropriately

  9. System Design • Eight components • Three categories • System specific monitors • Predictors • Viceroy/UI Figure 1. CABMAN system architecture

  10. System-specific Components • Detect various data from the OS • Battery status • By battery monitor • List and status of processes • By process monitor • Call logs • By call monitor • Context information to predict next charging opportunity • By context monitor • Separated from the OS to facilitate porting of CABMAN to the multiple platforms

  11. Charging Opportunity Predictor • Determine the charging opportunity is soon enough for battery • Should provide right information • Warn with high battery level if the charging opportunity is lowand vice versa • Use location sensing by GPS • To infer charging opportunity • Limited usage (still many devices don’t support) • Only respect to static charging opportunities

  12. Charging Opportunity Predictor • Cell based charging opportunity prediction algorithm • Used with following information • Location • Cell ID of connected phone • Chosen cells • Marked as being charging opportunities • Expected time to reach those cells • Prediction by pattern-matching against larger historical set of cell movement patterns • Current pattern is by using a number of samples being the current and most recent cell ids • Historical set is history of a number of days of cell movement patterns

  13. Charging Opportunity Predictor • Charging opportunity prediction algorithm • Based on current sample (ABC) • Search patterns including sample (DEABCFG) • between entry of the current cell and the next charging capable cell • Average time to provide prediction

  14. Call Time Predictor • Prevent other application to drain the battery for “crucial application” (telephony) • Three options • Ask the user to set a minimum call time level • Use past calling behavior to find the call time average • Find upper bound of used need of each day • Enhanced by compute weekdays and weekends separately • Hybrid approach • “keep twice my average call time available, and a minimum of 10 minutes for emergencies in addition to the predicted call time”

  15. Battery Lifetime Predictor • Monitor drain rate of the battery • Accurate estimation with same battery consumption level • But some are very over time • Different from battery age • Many don’t replace it • Propose a battery lifetime metric • Independent of battery age • Considering application’s battery usage

  16. Battery Lifetime Predictor Base curve of battery discharge Figure 2. A new laptop Figure 3. An old laptop Figure 4. HP iPAQ

  17. Battery Lifetime Predictor • Measure “discharge speedup factor” • Measure the battery capacity c1 and c2 at two time instances t1 and t2 with application running • Measure the battery capacity c1 and c2 at tow time instances t3 and t4 on idle state(base curve) • Calculated as (t4 –t3)/(t2-t1) • Divide the remaining lifetime of the battery by the discharge speedup factor to obtain the predicted remaining time for the battery

  18. Viceroy and User Interface • Continually monitor the battery lifetime prediction will expire before the next charging opportunity • If then, notify the user using the UI • When informed the user • Kill some battery-hungry applications • Make their behavior consume less power • Plan to charge device according to the timescale from the viceroy • Sacrifice crucial applications • If the user is at a place of charging opportunity • Ask user to charge directly

  19. Evaluation • Charging-opportunity predictor • Call-time predictor • Battery time predictor

  20. Charging Opportunity Predictor • History set of MIT’s Reality Mining project • 80 users, 9months • Varying parameters: sample size, history size • Increasing sample size generally increases accuracy and reliability • Sample size of 10 bottomed up • 40 days of historical data is optimal • User behavior changes • Average prediction error is 16%, 12 minutes

  21. Charging Opportunity Predictor Figure 4. Charging opportunity prediction error for various sample sizes and history sizes

  22. Call Time Predictor • Average prediction error is under a minuteout of the hour • Typical call is shot (90% are lees than 5 minutes) • Very few calls in a typical hour(75% with 2 calls or fewer) • Cannot predict a “long tail” • Occasionally long incoming calls • Try to preserve applications of telephony for emergencies, rendezvous

  23. Call Time Predictor Figure 5. Absolute call time prediction error for weekdays (top) and weekends (bottom)

  24. Call Time Predictor Figure 6. CDF of the length of phone calls (Left) and the number of calls made during each hour (Right)

  25. Battery-lifetime Predictor • Based on base curves • With new and old batteries • A set of applications • Web, music and video • By comparing • Actual consumption • Advanced Configuration and Power Interface (ACPI) • Estimation of the discharge speedup factor • Showed better prediction than ACPI

  26. Battery-lifetime Predictor Figure 7. Base curve together with discharge curves for the new HP laptop (Left) and old Dell laptop (Right)

  27. Battery-lifetime Predictor Table 1. comparing accuracy of algorithm with ACPI’s Figure 8. Base curve together with discharge curves (actual and derived) for HPiPAQ

  28. Discussion • Relatively more accurate prediction for average user whose life entropy is not very high. • Additional context information will be needed to improve the accuracy. • Calendar information, information about the travel plans of the user, charge-logs, etc.

  29. Conclusion • Describe three key components of CABMAN: • The use of context information such as location to predict the next charging opportunity • More accurate battery life prediction based on a discharge speedup factor • The notion of crucial applications such as telephony

  30. Conclusion • Evaluation Test results are very positive • Charging opportunity prediction exhibiting an average error of 12 minutes • Battery life prediction having average errors of between 4 and 12 minutes • Call time prediction algorithm has average errors measured in seconds • “minimum call time remaining”

  31. Pros (literature level) • 논문 구조가 복잡하지 않아 전반적으로 이해하기 쉬움 • 해결하고자 하는 문제가 이해하기 쉽게 설명됨

  32. Pros (System level) • Battery management를 위해 각각의 예측 알고리즘을 적용한 점이 돋보임 • Preliminary research 이기 때문에 각각의 predictor 및 시스템이 어떻게 구현될지는 구체적으로 알 수는 없지만, 일부분 (battery lifetime predictor)은 feasible 함 • 다른 predictor는 좀 더 feasible 해야 함 • 여러 모바일 기기에서 사용 가능하도록 system non-specific 한 접근이 돋보임

More Related