1University of Massachusetts,
Amherst
Users and Batteries : Interactions and Adaptive Power Management in
Mobile Systems
Nilanjan Banerjee1, Ahmad Rahmati2, Mark Corner1,
Sami Rollins3, Lin Zhong2
2 Rice University3University of San
Francisco
http://prisms.cs.umass.edu/llama
Scenario: why did my laptop switch of ?
You are riding a bus to work and you are five minutes away
you are working on your laptop finishing a presentation
Suddenly your laptop turns of ! Grrr … !!!
your laptop battery was running low
You would have charged your laptop within 5 minutes anyway
you could have completed your presentation
Scenario : working on an airplane
You are working on your presentation on a flight to Austria
Midway through your flight your laptop turns of
your battery could only last for three hours
Wish your laptop adapted to your charging behavior !
Problem : power management Vs user
Power management for mobile systems are not user-centric
do not adapt to changing user behavior and device modalities
No understanding of how users use energy of their mobile device
assumption: users desire maximum lifetime out of batteries
Battery
User
Solution: energy for the user
Understand user-battery interaction in mobile systems
when, why and where do users recharge
Built user-centric power management policy for mobile systems
policy which adapts to varying user-battery behavior
user behavior
energy management
OutlineUser-study on laptops and mobile phone
research methods for user-study
Insights from the user study
when, where, and why do users recharge batteries
how predictable are recharge patterns
User-centric power management
design and implementation, and evaluation of Llama
Related work
Conclusions
Study of user-battery interactionGoal : examine where, when, and why people recharge
subjects recruited from friends, family, mailing lists
used three complimentary research methods
10 Laptops10 Mobile phone age 20-26 years
10 Laptop 415 response10 Mobile phone 91 responses
56 Laptops 15-150 days10 Mobile phones 42-77 days
Trace Collection User Interviews In-situ survey
Trace collection Goal : collect quantitative records of battery level
Laptop implementation is Java based
runs on Microsoft Windows and Apple OS X
records measurements periodically
uploads data automatically to a central server once a day
Mobile phone tool is written in C++
runs on Microsoft Windows Mobile
tool distributed pre-installed on T-Mobile MDA phones
aggressive : wakes the phone very minute to take reading
User interviews
Gather qualitative data regarding user-battery interaction
understand context of recharge
Provided sample scenarios to participants to think about
last time the user was faced with a low battery condition ?
what impact did it have on their future behavior ?
Questions about when, why, and where users recharge ?
Encouraged users to tell their stories and anecdotes
In-situ pop-up survey
Filtered out intervals of less than 5 minutes between recharges
Disappears after a minute
Laptop
Mobile Phone
Goal: In-situ information about why users recharge
OutlineUser-study on laptops and mobile phone
research methods for user-study
Insights from the user study
when, where, and why do users recharge batteries
how predictable are recharge patterns
User-centric power management
design and implementation, and evaluation of Llama
Related work
Conclusions
Users have energy to spare
0
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0.16
0.18
0.2
3 13 23 33 43 53 63 73 83 93
Percentage at which battery is recharged
Pe
rce
nta
ge
of
rec
ha
rge
s Laptops
50% of the recharges occur when the battery is half full
Fraction of users use their laptops like desktops
Users have energy to spare
0
0.02
0.04
0.06
0.08
0.1
0.12
0.14
3 13 23 33 43 53 63 73 83 93
Percentage at which battery is recharges
Pe
rce
nta
ge
of
rec
ha
rge
s
Mobile Phones
60% of the recharges occur when the battery is half full
Most recharges occur between 25-75 %
Recharges are context driven
Convenient location
Convenient location
System Reminde
r
System Reminde
r
Convenient Time
Convenient Time
Low BatteryLow Battery
Limited Opportunitie
s Ahead
Limited Opportunitie
s Ahead
Laptops
Mobile Phones
Fraction of recharges are driven by context
Low battery corresponded to 40% of the battery remaining
Variations across users and devices
0
10
20
30
40
50
60
70
80
90
100
User 1 User 2 User 3 User 4 User 5
Bat
tery
Lev
el o
n R
echa
rge
Mobile Phones
Variation in recharge pattern across mobile phones and laptops
Variation across recharge patterns across users
Laptops
0
10
20
30
40
50
60
70
80
90
100
User 1 User 2 User 3 User 4 User 5
Ba
tte
ry L
ev
el
on
Re
ch
arg
e
Summary of the user-study
Recharges occur with significant energy remaining in batteries
Charging is mostly driven by context and battery levels
Users and devices show significant variation in battery usage
power management should adapt with users and devices
I always recharge every night
I usually charge in the office when the indicator shows 1 bar
User-centric power management
Users charge their system with significant battery left
accurately predict excess energy left in the battery
proactively use the remaining energy to improve QoS
Optimization framework for power management
maximize the excess energy usable by applications
minimize the probability of running out of battery
try to avoid true low battery levels
Llama : design and implementation
Example Scenario
Confidence of not exceeding battery capacity = 0.95
Llama determines present battery percentage (Cp) = 30%
creates a histogram of recharges below Cp (H)
Llama calculates 95% of the time user recharges by 10%
devote 10% to Llama application
Histogram ofRecharges
Histogram ofRecharges belowpresent capacity
Energy-adaptivealgorithm
Present BatteryCapacity
Energy forLlama app
Llama applications and deployment
Screen Brightness excess energy to adjust screen brightness
Web prefetching prefetching a random webpage download interval determines aggressiveness
Health monitoring reports preprogrammed data upload interval determines aggressiveness
Llama deployment demographics
Particulars Laptop Mobile Phone
Application Screen brightness Web prefetching
Health monitoring
Subjects 2 females , 8 male20-30 years
1 female, 9 males20-30 years
Number of Days 30 30
Llama evaluation
0
2
4
6
8
10
12
0 1 2 3 4 5 6 7 8 9
User ID
Pe
rce
nta
ge
Ba
tte
ry U
se
d b
y L
lam
a
0
5
10
15
20
25
0 1 2 3 4 5 6 7 8
User ID
Pe
rce
nta
ge
Ba
tte
ry U
se
d b
y L
lam
a
Laptops Mobile Phones
Llama used energy depending on battery left at recharge
Beneficial use of Llama more web data, and brighter display
Post-Llama recharge behavior
Particulars Laptop Mobile Phone
Number of recharges (per week)
Pre-Llama = 6.5Post-Llama = 7.8
Pre-Llama = 10.1Post-Llama = 8.9
Recharges below 5%
Pre-Llama = 1%Post-Llama = 1%
Pre-Llama = 4%Post-Llama = 7%
Feedback loop with user
User recharges at afixed percentage
Llama used upexcess energy
Recharge cyclebecomes shorter
Recharge cycle becomes shorter and shorter, frustrating the user
Plan to address the problem in future versions of Llama
Post-Llama user study
Interviews to evaluate negative effects of Llama
impact of Llama on battery lifetime
All mobile phone users but one showed similar satisfaction
“The battery lifetime was better last month, I have to recharge it every day now, but it used to be every day and a half”
It must have been small, since I didn’t notice it
Even though I didn’t notice it, I would definitely care in situations where I require maximum battery life
Laptop user
Future work
Evaluate the positive effects of Llama
what are the user-perceived benefits of Llama ?
Improve the prediction algorithm of Llama
use contextual information such as location, work patterns
Experiment on different mobile devices like music players
less biased or demographically weighted subject selection
Related workMyExperience in-situ survey tool [Mobisys 2007]
tool for in-situ profiling and survey
Human factor in energy management
user-interface design on energy efficiency [Vallero et al.]
visual perception to reduce energy of LCDs [Chen et al.]
Tools for studying mobile users in natural settings
logging tool for studying HCI [Demumieux et al.]
Balance performance and system-wide energy consumption
Odyssey [Flinn et al.], Ecosystem [Zeng et al.]
Conclusions
First glimpse of user-battery interaction
traces would be available through the traces.cs project
User study produced three key observations
users leave excess energy in the battery on recharge
charging behavior is driven by opportunity and context
significant variations across users and systems
Built an user-centric energy management system called Llama
it can scale energy usage to user behavior
1University of Massachusetts,
Amherst
Users and Batteries : Interactions and Adaptive Power Management in
Mobile Systems
Nilanjan Banerjee1, Ahmad Rahmati2, Mark Corner1,
Sami Rollins3, Lin Zhong2
2 Rice University3University of San
Francisco
http://prisms.cs.umass.edu/llama
HotMobile 2008
Napa, CA, February 25-26, 2008Submissions: October 16, 2007
Napa, CA, February 25-26, 2008Submissions: October 16, 2007