User-Aware Power Management
for Mobile Devices
8/11/2016 9:38 AM
Geunsik Lim, Changwoo Min, Dong Hyun Kang, and Young Ik Eom
Sungkyunkwan University, Korea
Samsung Electronics Co., Ltd., Korea
IEEE GCCE 2013

2/15
Introduction
Breakdown of power consumption in mobile
Our proposal
 User-aware power management
Evaluation
Conclusion
Outline

3/15
Battery Monitor Powertutor AppScope JuiceDefender
Physical Extension External Battery Solar Recharger
How to extend battery lifeHardwareSoftware

4/15
1. Turn off radios like WiFi, Bluetooth etc : If you don’t ne
ed them, turn them off. You don’t need GPS or Bluetooth
all the time.
2. Turn off apps that use more battery: You can close the
applications (e.g. running applications, sync) you don’t ne
ed to supply battery.
3. Decrease Screen Brightness & Screen timeout: You can
decrease brightness to save battery little more
4. Turn on Airplane Mode : Airplane Mode is useful if you
don’t need to connect to WiFi or mobile data networks.
Extending battery life by users

5/15
• Typical pattern of power consumption of a mobile device without running
any user applications (factory reset).
• The breakdown of each factor about degradation of battery lifespan.
• Platform developers does not report no cases of battery degradation.
* Experimental Device: Galaxy S2 (1,650mAh)
Breakdown of power consumption in mobile

6/15
How should “Our idea” …
1. Recognize the idle time of users?
2. Execute wake-up automatically from shutdown status?
3. How can user control?
Challenges: a great way to get new insights

7/15
• Our approach is a system that consists of the RTC-based kernel feature and
the user-space mobile app to extend battery life.
• The major goal of Our system is to control the turn-off and reboot
operation as well as suspend-to-ram/suspend-to-disk during the sleep
period of customers.
After
1 day
After
2 days
After
3 days
After
0 day
User-Aware Power
Management for
Mobile Devices!!!
Problem Create idea How
What is our idea?

8/15
Hardware
Kernel-space
(4)Battery
Timer
(3)SleepLevel
Controller
(2) Sleep Time Manager
(1) User-space Client
Memory
Suspend-To-RAM
Suspend-To-Disk
CompleteOff
Overall diagram of proposed system 1/2
Time Scheduling
Based on RTC-Timer
User-space Client
START
Sleep Time Manager Suspend-To-RAM (S3)
Suspend-To-Disk
Turn-Off
Power Saving
Level == {1|2|3}
Wake-up (By Battery Timer)
Service Restoration END
(1)
(2)
(3)
OverallArchitectureEntireFlowDiagram

9/15
Suspend-To-RAM
Suspend-To-Disk
Complete Off
Sleep
Executor
TimeWakeup
H/W
Timer
Auto
Wake-up
Saving Time
(Time to reduce power
consumption)
Time
Checker
User space
Application
Middle ware
User space
Application
Energy
Miser Client
User space
Application
User space
Application
User-spaceKernel-space
Sleep
(4) Battery
Timer
(3) Sleep Level Controller
(2) Sleep Time Manager
(1) User-space Client
Memory
Overall diagram of proposed system 2/2

10/15
Sleep()
/sys/…/current_clocksource
UserKernelHardware
gettimeofday() clock_* API nanosleep()
do_gettimeofday() xtime HRT (High Resolution Timer)
Clocksource
Framework
Clockevent
Framework
APICJiffies ACPI TSC
PIT
HPET
ITimer API
…
RTC
Overall structure of RTC-based H/W timer
• Timekeeping using clocksource and clockevents based on RTC timer.

11/15
• User-aware UI interface for complete automation.
Suspend-to-
resume
Suspend-to-
disk
Complete-off
By
interrupt
30 Min.
1 Hour
30 Min.
1 Hour
2 Hour
Customi
ze
Customi
ze
Wakeup Time Sleep Time
Check the period of display-off
status “Wakeup Time”
If display-off period happens
again, repeat the execution of
“Wakeup time/Sleep time”
If the display-off period is bigger
than user-setting, Execute one of
the three methods “Sleep Level”.
Restart services after “Sleep Time”.
Start
End
Sleep Level
Relation between user-space client and operation

12/15
Because of turn-on of some peripheral
devices (Just, Engineering issue)
Evaluation: w/o additional user apps
• Galaxy S2: 1,650mAh
• Galaxy S3: 2,100mAh
• Nexus 7: 4,325mAh
• Laptop : 4,400mAh

13/15
*Active devices (WiFi, Bluetooth, GPS)
*Sync (Gmail, Facebook, etc)
*Background daemons (Skype, Kakao)
• Galaxy S2: 1,650mAh
• Galaxy S3: 2,100mAh
• Nexus 7: 4,325mAh
• Laptop : 4,400mAh
Evaluation: w/ additional user apps

14/15
• Even when users do not use any applications, battery lifespan
decreases continually. e.g., (1) device operations & service
daemons and (2) application daemons
• A new power management system that controls the power
supply completely via four major components such as (1) user-
space client, (2) sleep time manager, (3) sleep level controller,
and (4) battery timer.
• Our system extends the battery lifespan compared to the existing
system.
a. w/o additional user applications: power saving is 18%
b. w/ additional user applications: power saving is 34%
• In addition, proposed system reschedules automatically all
services in advance before the user tries to reuse the mobile
device.
Conclusion

15/15
Thank you for your attention.
Any questions?

16/15
1. Who cares about Timer-based phone. Can you do it
for Window and iPhone?
2. Are you one of those who don’t care about the
power consumption for restarting of service
daemons?
3. Sounds too good. Are there any limitations?
4. Are you going to release it? Or is it a one of paper?
5. I totally don’t get why you are doing this?
FAQ

17/15
1. This approach is Hardware-based software solution.
But, the most of mobile devices have H/W alarm
timer.
2. Users needs to do battery consumption to relaunch
the services of mobile platform. (Up to 1%)
3. If you can always use battery adapter in your real life?
We only focus on the battery lifespan in the mobility
view.
Limitation

18/15
BACKUP SLIDES
In Case We Have More Time…

19/15
Related Work: Power Tutor
• Power tutor
• Mian Dong and Lin Zhong, “Self-constructive, high-rate energy modeling
for battery-powered mobile systems,” in Proc. of the ACM/USENIX Int.
Conf. Mobile Systems, Applications, and Services (MobiSys), June 2011.
Architecture of Sesame with three major components: data collector,
model constructor and model manager

20/15
Related Work: AppScope
• AppScope
• C. Yoon, D. Kim, W. Jung, C. Kang, and H. Cha, “Appscope: Application
energy metering framework for android smartphone using kernel activity
monitoring,” in Proc. of the USENIX Annual Technical Conference
(USENIX ATC), s2012.
AppScope overview on Android platform

21/15
Related Work: Power Scheduling Algorithm
• R. Nathuji and K. Schwan, “Reducing system level power consumption for mobile
and embedded platforms,” in Proc. of the ARCS'05 Proceedings of the 18th
international conference on Architecture of Computing Systems conference on
Systems Aspects in Organic and Pervasive Computing, pp. 18-32, 2005.
• A new process scheduling algorithm which accumulates device usage information
in the form of device windows to make power a first class resource
Bursting Device Accesses

22/15
Related Work: Energy-aware application Performance
• T. Simunic, L. Benini, P. Glynn, and G. D. Micheli, “Dynamic power
management for portable systems,” in Proc. of the 6th annual
international conference on Mobile computing and networking
(MobiCom),” pp. 11-19. 2000.
• The time-indexed SMDP model (TISMDP) that we use to derive optimal
policy for DPM(Dynamic Power Management) in portable systems.

23/15
Related Work: DFS-based Power Saving Algorithm
The operating frequency of CPU is dynamically changed in accordance to the estimated processing
burden of each video frame. The frame-based dynamic frequency scaling is applied to the MPEG4 v
ideo decoding in the portable kitchen TV system, (e.g., ARM926EJ-S Processor)
Summary: Dynamic Frequency Scaling Based Power Saving Algorithm for a
Portable Kitchen TV Won-Jong Kim (Chung-Ang University, Korea)

24/15
Related Work: Accurate GPU Power Estimation
GPU Applications, GPU Power Consumption Analysis, GPU Power
Previous works: utilization based, system call based <-- online pow
er profiling techniques
The power profile obtained from each technique with that obtained
from a hardware power monitoring device.
To calculate accuracy, we compare the power profile obtained from
each technique with that obtained from a hardware power monitori
ng device.
Summary: Accurate GPU Power Estimation for Mobile Device Power Profiling
Minyong Kim (Korea University, Korea)

25/15
Related Work: Battery Manager (by Google)
Battery Manager related JAVA APIs
http://developer.android.com/reference/android/os/BatteryManager.html
Monitoring the Battery Level and Charging State
http://developer.android.com/training/monitoring-device-state/battery-monitoring.html
Optimizing Battery Life
http://developer.android.com/training/monitoring-device-state/index.html

26/15
E N D

27/15
15 minutes for oral presentations will include:
1. A brief introduction of the speaker by Session Chair
2. The body of your presentation
3. A brief question-and-answer period
4. A setting minute to change over to the next presenter
Please prepare the presenter's brief bio and give it to Session Chairs before
beginning of your session.
Each presentation room has a Windows PC with a PowerPoint Viewer and an
Adobe Reader. You may also use your own Windows PC or Macintosh via a
VGA (DSUB 15-pin) connector.
Attn: Best Paper Award and Best Student Paper Award will be chosen
from papers presented at Award and Student Award Candidate Sessions on
Oct. 1st. The award winners will be announced at the Award Banquet on Oct.
3rd.
Oral Presentation Guideline