Upload
others
View
0
Download
0
Embed Size (px)
Citation preview
Diversity in Smartphone Energy Consumption
Earl Oliver
David R. Cheriton School of Computer Science
University of Waterloo
1
Tuesday, September 14, 2010
Take-aways
Built a comprehensive dataset that contains over one millennium of energy consumption behaviour from over 17300 BlackBerry users.
2
Tuesday, September 14, 2010
Take-aways
Built a comprehensive dataset that contains over one millennium of energy consumption behaviour from over 17300 BlackBerry users.
Report on the characteristics of three distinct smartphone user types:
2
Tuesday, September 14, 2010
Take-aways
Built a comprehensive dataset that contains over one millennium of energy consumption behaviour from over 17300 BlackBerry users.
Report on the characteristics of three distinct smartphone user types:
opportunistic chargers
2
Tuesday, September 14, 2010
Take-aways
Built a comprehensive dataset that contains over one millennium of energy consumption behaviour from over 17300 BlackBerry users.
Report on the characteristics of three distinct smartphone user types:
opportunistic chargers
light-consumers
2
Tuesday, September 14, 2010
Take-aways
Built a comprehensive dataset that contains over one millennium of energy consumption behaviour from over 17300 BlackBerry users.
Report on the characteristics of three distinct smartphone user types:
opportunistic chargers
light-consumers
nighttime chargers
2
Tuesday, September 14, 2010
Motivation
Two dominant trends
Mobile applications
3
Battery energydensity
Tuesday, September 14, 2010
Execution continuum
Energy conversion rate
GSM phone calls < 2W
Duration of use Video playback
~1.1 W
Audio~50 mW
Video recording~1.4 W
GPS navigation~620 mW
Bluetooth, logging tools
< 2.5 mW
Offline tasks: e-mail,
calendar, etc.< 10 mW
Web browsing
(WiFi)~800 mW
The next frontier in mobile applications:MyTube, PocketBay, etc.
Media indexingservice
4
Tuesday, September 14, 2010
Execution continuum
Energy conversion rate
GSM phone calls < 2W
Duration of use Video playback
~1.1 W
Audio~50 mW
Video recording~1.4 W
GPS navigation~620 mW
Execution autonomy
Bluetooth, logging tools
< 2.5 mW
Offline tasks: e-mail,
calendar, etc.< 10 mW
Web browsing
(WiFi)~800 mW
The next frontier in mobile applications:MyTube, PocketBay, etc.
Media indexingservice
4
Tuesday, September 14, 2010
Execution continuum
Energy conversion rate
GSM phone calls < 2W
Duration of use Video playback
~1.1 W
Audio~50 mW
Video recording~1.4 W
GPS navigation~620 mW
Execution autonomy
Execution requires explicit user interactionBluetooth, logging tools
< 2.5 mW
Offline tasks: e-mail,
calendar, etc.< 10 mW
Web browsing
(WiFi)~800 mW
The next frontier in mobile applications:MyTube, PocketBay, etc.
Media indexingservice
4
Tuesday, September 14, 2010
Execution continuum
Energy conversion rate
GSM phone calls < 2W
Duration of use Video playback
~1.1 W
Audio~50 mW
Video recording~1.4 W
GPS navigation~620 mW
Execution autonomy
Execution requires some user interactionBluetooth,
logging tools< 2.5 mW
Offline tasks: e-mail,
calendar, etc.< 10 mW
Web browsing
(WiFi)~800 mW
The next frontier in mobile applications:MyTube, PocketBay, etc.
Media indexingservice
4
Tuesday, September 14, 2010
Execution continuum
Energy conversion rate
GSM phone calls < 2W
Duration of use Video playback
~1.1 W
Audio~50 mW
Video recording~1.4 W
GPS navigation~620 mW
Execution autonomy
Execution requires little or no user interaction
Bluetooth, logging tools
< 2.5 mW
Offline tasks: e-mail,
calendar, etc.< 10 mW
Web browsing
(WiFi)~800 mW
The next frontier in mobile applications:MyTube, PocketBay, etc.
Media indexingservice
4
Tuesday, September 14, 2010
Execution continuum
Energy conversion rate
GSM phone calls < 2W
Duration of use Video playback
~1.1 W
Audio~50 mW
Video recording~1.4 W
GPS navigation~620 mW
Execution autonomy
Execution requires little or no user interaction
Bluetooth, logging tools
< 2.5 mW
Offline tasks: e-mail,
calendar, etc.< 10 mW
Web browsing
(WiFi)~800 mW
The next frontier in mobile applications:MyTube, PocketBay, etc.
Media indexingservice
4
Tuesday, September 14, 2010
Execution continuum
Energy conversion rate
GSM phone calls < 2W
Duration of use Video playback
~1.1 W
Audio~50 mW
Video recording~1.4 W
GPS navigation~620 mW
Execution autonomy
Execution requires little or no user interaction
Bluetooth, logging tools
< 2.5 mW
Offline tasks: e-mail,
calendar, etc.< 10 mW
Web browsing
(WiFi)~800 mW
The next frontier in mobile applications:MyTube, PocketBay, etc.
Media indexingservice
4
Tuesday, September 14, 2010
Wireless opportunistic communication
The next frontier in mobile applications
5
Tuesday, September 14, 2010
Wireless opportunistic communication
requires autonomous execution
The next frontier in mobile applications
5
Tuesday, September 14, 2010
Wireless opportunistic communication
requires autonomous execution
Continuous execution maximize connection opportunities
The next frontier in mobile applications
5
Tuesday, September 14, 2010
Wireless opportunistic communication
requires autonomous execution
Continuous execution maximize connection opportunities
Unfettered energy consumption can deplete finite energy resources
The next frontier in mobile applications
5
Tuesday, September 14, 2010
Wireless opportunistic communication
requires autonomous execution
Continuous execution maximize connection opportunities
Unfettered energy consumption can deplete finite energy resources
Unacceptable
The next frontier in mobile applications
5
Tuesday, September 14, 2010
Execution continuum
Energy conversion rate
Duration of use
Execution autonomy
Execution requires little or no user interaction
How to maximize benefit while minimizing impact?
The next frontier in mobile applications:MyTube, PocketBay, etc.
6
Tuesday, September 14, 2010
Smartphone usage study
Need to understand how smartphone users consume energy on their personal mobile devices.
7
Tuesday, September 14, 2010
Smartphone usage study
Need to understand how smartphone users consume energy on their personal mobile devices.
Metrics:
7
Tuesday, September 14, 2010
Smartphone usage study
Need to understand how smartphone users consume energy on their personal mobile devices.
Metrics:
charge/discharge duration
7
Tuesday, September 14, 2010
Smartphone usage study
Need to understand how smartphone users consume energy on their personal mobile devices.
Metrics:
charge/discharge duration
charge/discharge initiation time and battery level
7
Tuesday, September 14, 2010
Smartphone usage study
Need to understand how smartphone users consume energy on their personal mobile devices.
Metrics:
charge/discharge duration
charge/discharge initiation time and battery level
battery level throughout the day
7
Tuesday, September 14, 2010
Smartphone usage study
Need to understand how smartphone users consume energy on their personal mobile devices.
Metrics:
charge/discharge duration
charge/discharge initiation time and battery level
battery level throughout the day
charge/discharge rate
7
Tuesday, September 14, 2010
Charge duration
0
20
40
60
80
100
1 10 100 1000 10000 100000
CD
F (%
)
Charge duration (seconds)
Global Charge Duration
Opportunistic chargerLight-consumers
Nighttime chargers
Tuesday, September 14, 2010
Discharge duration
0
20
40
60
80
100
1 10 100 1000 10000 100000
CD
F (%
)
Time between charges (seconds)
Global Discharge Duration
Opportunistic chargerLight-consumers
Nighttime chargers
Tuesday, September 14, 2010
Charge initiation time
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
1.8
Sunday Monday Tuesday Wednesday Thursday Friday Saturday
Tota
l sam
ples
(per
cent
)
Days of the week (1800 second sample interval)
Weekly Smartphone Charge Initiation Time (normalized to GMT)
Opportunistic chargerLight-consumers
Nighttime chargers
Tuesday, September 14, 2010
Charge initiation level
0
20
40
60
80
100
10 20 30 40 50 60 70 80 90
CD
F (%
)
Battery level at charge (%)
Global Battery Level at Charge
Opportunistic chargerLight-consumers
Nighttime chargers
Tuesday, September 14, 2010
Daily battery level
0
20
40
60
80
100
Sunday Monday Tuesday Wednesday Thursday Friday Saturday
Batte
ry le
vel (
perc
ent)
Days of the week (1800 second sample interval)
Daily Battery Level (normalized to GMT)
Opportunistic chargerLight-consumers
Nighttime chargers
Tuesday, September 14, 2010
Discharge rate
1.5
2
2.5
3
3.5
4
4.5
5
5.5
Sunday Monday Tuesday Wednesday Thursday Friday Saturday
Batte
ry d
isch
arge
rate
(bat
tery
leve
l / h
our)
Days of the week (1800 second sample interval)
Daily Battery Discharge Rate (normalized to GMT)
Opportunistic chargerLight-consumers
Nighttime chargers
Tuesday, September 14, 2010
Summary
Comprehensive user study
opportunistic chargers
light-consumers
nighttime chargers)
Tuesday, September 14, 2010
Summary
Comprehensive user study
opportunistic chargers
light-consumers
nighttime chargers)
Foundation for:
Regulating mobile application execution based on a predicted future battery level.
Predicting the successful execution rate of energy intensive applications given known energy consumption behaviour.
Tuesday, September 14, 2010
User classification
Cluster by device
Cluster by energy consumption/replenishment characteristics
Which set of characteristics are best?
The set that yield the best prediction results.
Select clustering
parameters
Cluster k-means
Apply predictor (3-way cross
validation)
Retrieve mean prediction error
rate
Parameter analysis
16
Tuesday, September 14, 2010
User classification
Cluster by device
Cluster by energy consumption/replenishment characteristics
17
Tuesday, September 14, 2010
User classification
Cluster by device
Cluster by energy consumption/replenishment characteristics
Which set of characteristics are best?
17
Tuesday, September 14, 2010
User classification
Cluster by device
Cluster by energy consumption/replenishment characteristics
Which set of characteristics are best?
The set that yield the best prediction results.
17
Tuesday, September 14, 2010
Prediction algorithmδcharget = mean charge duration of a charge initiated during bucket(t)
δdischarget = mean discharge duration of a discharge initiated during bucket(t)
ρcharget = mean charge rate during bucket(t)
ρdischarget = mean charge rate during bucket(t) bucket size = 30 minutes
tlast = last time that the device began charging or discharging
18
Tuesday, September 14, 2010
Prediction algorithm
Time
Batterylevel
100%
0%
δcharget = mean charge duration of a charge initiated during bucket(t)
δdischarget = mean discharge duration of a discharge initiated during bucket(t)
ρcharget = mean charge rate during bucket(t)
ρdischarget = mean charge rate during bucket(t) bucket size = 30 minutes
tlast = last time that the device began charging or discharging
18
Tuesday, September 14, 2010
Prediction algorithm
Time
Batterylevel
100%
0%
True battery level
δcharget = mean charge duration of a charge initiated during bucket(t)
δdischarget = mean discharge duration of a discharge initiated during bucket(t)
ρcharget = mean charge rate during bucket(t)
ρdischarget = mean charge rate during bucket(t) bucket size = 30 minutes
tlast = last time that the device began charging or discharging
18
Tuesday, September 14, 2010
Prediction algorithm
Time
Batterylevel
100%
0%
True battery level
t
δcharget = mean charge duration of a charge initiated during bucket(t)
δdischarget = mean discharge duration of a discharge initiated during bucket(t)
ρcharget = mean charge rate during bucket(t)
ρdischarget = mean charge rate during bucket(t) bucket size = 30 minutes
tlast = last time that the device began charging or discharging
18
Tuesday, September 14, 2010
Prediction algorithm
Time
Batterylevel
100%
0%
True battery level
t t + 1 hour
δcharget = mean charge duration of a charge initiated during bucket(t)
δdischarget = mean discharge duration of a discharge initiated during bucket(t)
ρcharget = mean charge rate during bucket(t)
ρdischarget = mean charge rate during bucket(t) bucket size = 30 minutes
tlast = last time that the device began charging or discharging
18
Tuesday, September 14, 2010
Prediction algorithm
Time
Batterylevel
100%
0%
True battery level
t t + 1 hour
δcharget = mean charge duration of a charge initiated during bucket(t)
δdischarget = mean discharge duration of a discharge initiated during bucket(t)
ρcharget = mean charge rate during bucket(t)
ρdischarget = mean charge rate during bucket(t) bucket size = 30 minutes
tlast = last time that the device began charging or discharging
tlast
18
Tuesday, September 14, 2010
Prediction algorithm
Time
Batterylevel
100%
0%
True battery level
t t + 1 hour
δdischargetlast
δcharget = mean charge duration of a charge initiated during bucket(t)
δdischarget = mean discharge duration of a discharge initiated during bucket(t)
ρcharget = mean charge rate during bucket(t)
ρdischarget = mean charge rate during bucket(t) bucket size = 30 minutes
tlast = last time that the device began charging or discharging
tlast
18
Tuesday, September 14, 2010
Prediction algorithm
Time
Batterylevel
100%
0%
True battery level
t t + 1 hour
δdischargetlast
ρdischarget
δcharget = mean charge duration of a charge initiated during bucket(t)
δdischarget = mean discharge duration of a discharge initiated during bucket(t)
ρcharget = mean charge rate during bucket(t)
ρdischarget = mean charge rate during bucket(t) bucket size = 30 minutes
tlast = last time that the device began charging or discharging
tlast
18
Tuesday, September 14, 2010
Prediction algorithm
Time
Batterylevel
100%
0%
True battery level
t t + 1 hour
δdischargetlast
ρdischargetρdischarget+1
δcharget = mean charge duration of a charge initiated during bucket(t)
δdischarget = mean discharge duration of a discharge initiated during bucket(t)
ρcharget = mean charge rate during bucket(t)
ρdischarget = mean charge rate during bucket(t) bucket size = 30 minutes
tlast = last time that the device began charging or discharging
tlast
18
Tuesday, September 14, 2010
Prediction algorithm
Time
Batterylevel
100%
0%
True battery level
t t + 1 hour
δdischargetlast
ρdischargetρdischarget+1
ρdischarget+2
δcharget = mean charge duration of a charge initiated during bucket(t)
δdischarget = mean discharge duration of a discharge initiated during bucket(t)
ρcharget = mean charge rate during bucket(t)
ρdischarget = mean charge rate during bucket(t) bucket size = 30 minutes
tlast = last time that the device began charging or discharging
tlast
18
Tuesday, September 14, 2010
Prediction algorithm
Time
Batterylevel
100%
0%
True battery level
t t + 1 hour
δdischargetlast
ρdischargetρdischarget+1
ρdischarget+2
δcharget = mean charge duration of a charge initiated during bucket(t)
δdischarget = mean discharge duration of a discharge initiated during bucket(t)
ρcharget = mean charge rate during bucket(t)
ρdischarget = mean charge rate during bucket(t) bucket size = 30 minutes
tlast = last time that the device began charging or discharging
tlast
Predictedbattery level
Predictionerror
18
Tuesday, September 14, 2010
Prediction algorithm
Time
Batterylevel
100%
0%
t1
δcharget = mean charge duration of a charge initiated during bucket(t)
δdischarget = mean discharge duration of a discharge initiated during bucket(t)
ρcharget = mean charge rate during bucket(t)
ρdischarget = mean charge rate during bucket(t) bucket size = 30 minutes
tlast = last time that the device began charging or discharging
True battery level
19
Tuesday, September 14, 2010
Prediction algorithm
Time
Batterylevel
100%
0%
t1 t1 +6 hours
δcharget = mean charge duration of a charge initiated during bucket(t)
δdischarget = mean discharge duration of a discharge initiated during bucket(t)
ρcharget = mean charge rate during bucket(t)
ρdischarget = mean charge rate during bucket(t) bucket size = 30 minutes
tlast = last time that the device began charging or discharging
True battery level
19
Tuesday, September 14, 2010
Prediction algorithm
Time
Batterylevel
100%
0%
t1 t1 +6 hours
tlast
δdischargetlast
δcharget = mean charge duration of a charge initiated during bucket(t)
δdischarget = mean discharge duration of a discharge initiated during bucket(t)
ρcharget = mean charge rate during bucket(t)
ρdischarget = mean charge rate during bucket(t) bucket size = 30 minutes
tlast = last time that the device began charging or discharging
True battery level
19
Tuesday, September 14, 2010
Prediction algorithm
Time
Batterylevel
100%
0%
t1 t1 +6 hours
tlast
δdischargetlast
δcharget = mean charge duration of a charge initiated during bucket(t)
δdischarget = mean discharge duration of a discharge initiated during bucket(t)
ρcharget = mean charge rate during bucket(t)
ρdischarget = mean charge rate during bucket(t) bucket size = 30 minutes
tlast = last time that the device began charging or discharging
True battery level
t2
19
Tuesday, September 14, 2010
Prediction algorithm
Time
Batterylevel
100%
0%
t1 t1 +6 hours
tlast
δdischargetlast
δcharget = mean charge duration of a charge initiated during bucket(t)
δdischarget = mean discharge duration of a discharge initiated during bucket(t)
ρcharget = mean charge rate during bucket(t)
ρdischarget = mean charge rate during bucket(t) bucket size = 30 minutes
tlast = last time that the device began charging or discharging
True battery level
t2
δcharget2
19
Tuesday, September 14, 2010
Prediction algorithm
Time
Batterylevel
100%
0%
t1 t1 +6 hours
tlast
δdischargetlast
δcharget = mean charge duration of a charge initiated during bucket(t)
δdischarget = mean discharge duration of a discharge initiated during bucket(t)
ρcharget = mean charge rate during bucket(t)
ρdischarget = mean charge rate during bucket(t) bucket size = 30 minutes
tlast = last time that the device began charging or discharging
True battery level
t2
ρcharget2
δcharget2
19
Tuesday, September 14, 2010
Prediction algorithm
Time
Batterylevel
100%
0%
t1 t1 +6 hours
tlast
δdischargetlast
δcharget = mean charge duration of a charge initiated during bucket(t)
δdischarget = mean discharge duration of a discharge initiated during bucket(t)
ρcharget = mean charge rate during bucket(t)
ρdischarget = mean charge rate during bucket(t) bucket size = 30 minutes
tlast = last time that the device began charging or discharging
True battery level
t2
ρcharget2
δcharget2
ρcharget2+1
19
Tuesday, September 14, 2010
Prediction algorithm
Time
Batterylevel
100%
0%
t1 t1 +6 hours
tlast
δdischargetlast
δcharget = mean charge duration of a charge initiated during bucket(t)
δdischarget = mean discharge duration of a discharge initiated during bucket(t)
ρcharget = mean charge rate during bucket(t)
ρdischarget = mean charge rate during bucket(t) bucket size = 30 minutes
tlast = last time that the device began charging or discharging
True battery level
t2
ρcharget2
δcharget2
t3
δdischarget3
ρcharget2+1
t4
δcharget4
19
Tuesday, September 14, 2010
Prediction algorithm
Time
Batterylevel
100%
0%
t1 t1 +6 hours
tlast
δdischargetlast
δcharget = mean charge duration of a charge initiated during bucket(t)
δdischarget = mean discharge duration of a discharge initiated during bucket(t)
ρcharget = mean charge rate during bucket(t)
ρdischarget = mean charge rate during bucket(t) bucket size = 30 minutes
tlast = last time that the device began charging or discharging
True battery level
t2
ρcharget2
δcharget2
t3
δdischarget3
ρcharget2+1
t4
δcharget4
Predictionerror
19
Tuesday, September 14, 2010