pada: power-aware development assistant of mobile
TRANSCRIPT
PADA: Power-aware Development Assistantfor Mobile Sensing Applications
Chulhong Min1, Seungchul Lee1, Changhun Lee1, Youngki Lee2,Seungwoo Kang3, Seungpyo Choi1, Wonjung Kim1, Junehwa Song1
1KAIST, 2Singapore Management University, 3KOREATECH
UbiComp 2016
How do you optimize the power use?
Power-aware Development
Extremely BURDENSOMEdue to REPETITIVE power evaluation
is Difficult!!
Mobile sensing applicationsmake it more challenging!!
Suppose You DevelopTransport Mode Tracker
Power should be evaluatedin REAL-LIFE situations, REPETITIVELY!!
19 22
5
20s 30s 40s
Age
12
6
14
7 7
0-2 2-4 4-6 6-8 > 8
Development experiences (years)
13
23
10
Expert Intermediate Beginner
Self-reported Skill
61%
78%
Survey of 46 professional mobile developers
Do you measure the power use of an application?
No35%
Yes65%
35% do not measure
When do you start considering the energy efficiency?
No35%
After full impl.35%
Before full impl.
30%35% do not measure
35% measure after full implementation
How do you or your team measure the power? (Check all that apply)
8
31
15
5
1
9 8
40
21
# of
par
ticip
ants
QA Developer
Battery levelH/W usageobservation
Power monitor
I don’t know
No
Other
Rank the power cost ofscreen, GPS, storage, and accelerometer
25%
69%
6%
0% 10% 20% 30% 40% 50% 60% 70% 80%
Correct
Incorrect
I don't know
Key Takeaways
Mobile developers mostly
• Perceive the energy efficiency very important
• Often fail to consider the power
• Use inconvenient, inaccurate methods
• Have limited understanding on energy use
PADA: Power-aware Development Assistant for Mobile Sensing Applications
Facilitating REPETITIVE power evaluation
Supporting various REAL-LIFE situations
Providing ENRICHED power information
http://static6.businessinsider.com/image/554ce87e69bedd9438d97cdd-1200-924/happy-programmer-work.jpg
PADA Architecture
PADAfront end
Emulationmanager
Power emulatorPower emulatorPower emulatorPower emulator
Context-labeled sensor trace
App executables
Evaluation settings
Sensor Trace Collector
Power analyzer
APIs for power evalutation
Request
Power analysis result
Developer
Approach: Context-driven Record&Replay
App execution
H/W usage tracking
Sensor trace
Output: power
Record
Replay
Power Emulator
SensorManager
BT/WiFiManager
LocationManager
Sensing app
AlarmManager
PowerManager
sensor data
Kernel
Andr
oid
Fram
ewor
k
wake-up msg.
H/W Usage
Monitor
Sensor Emulator traceAccel.Accel. BT Wi-Fi GPS
PowerEstimator
Power Evaluation RequestWeb Demo: http://pada-web.github.io
Upload APK file
Select scenarios to test
In-depth AnalysisWeb Demo: http://pada-web.github.io
overview
power consumption over time
In-depth AnalysisWeb Demo: http://pada-web.github.io
hardware use pattern
In-depth AnalysisWeb Demo: http://pada-web.github.io
context and custom logs
Comparative AnalysisWeb Demo: http://pada-web.github.io
Location interval10sec? 20sec? 30sec?
<parameter.xml><parameter type=“location_interval”>
<value>10</value><value>20</value><value>30</value>
</parameter>
<application code>interval = PADA.getParameterValue(
“location_interval”, “parameter.xml”);requestLocation(GPS_PROVIDER, interval, 0, this);
Comparative AnalysisWeb Demo: http://pada-web.github.io
Power-aware implementation Power Tuning
Power-aware Implementation:Experimental Setup
Goal: implement iTrackwhile meeting power requirements
14 programmers
3 hour-longdevelopment
iTrack: energy-efficient location tracker
# Traces [minutes, context] Power requirements
1
Out
door [5, staying] DutyCPU ≤ 20%, # of BT/GPS calls = 0, …
2 [5, walking] # of BT calls = 1, # of GPS calls = 5, …
3 [3, walking], [2, staying] DutyACC ≤ 15%, # of alarm calls ≤ 50, …
4
Indo
or
[5, staying] DutyCPU ≤ 20%, # of BT/GPS calls = 0, …
5 [5, walking] TimeBT ≤ 120s, # of GPS calls = 0, …
6 [3, walking], [2, staying] DutyCPU ≤ 40%, # of BT calls = 3, …
Power-aware Implementation:Key Findings
Lowering burden of power evaluation
Repeatability of power evaluation
Logical bug detection
“While using PADA, I didn’t need to go out for testing GPS and I really liked it.” [P7]
Power-aware Implementation:Key Findings
Lowering burden of power evaluation
Repeatability of power evaluation
Logical bug detection
“When I saw the power report of my modified application, I realized that my modification was really bad. Without PADA, it could be harder to notice this mistake” [P9]
Power-aware Implementation:Key Findings
Lowering burden of power evaluation
Repeatability of power evaluation
Logical bug detection
Two alarm services for periodic sensing of accelerometer and Bluetooth with the same identifier [P12]
Power Tuning:Experimental Setup
SocioHotspot:Social hotspot tracker with BT scanning
Goal: tune SocioHotspot to balance energy efficiency and accuracy
5 programmers
select BT scan intervalwith/without PADA
Power Tuning without PADA:Key Findings
Inconsistency of daily behavior
Nontrivial burden
Limited result
“I tried to move around similar to yesterday, but it was practically infeasible.” [PB]
Power Tuning without PADA:Key Findings
Inconsistency of daily behavior
Nontrivial burden
Limited result
“It may need ten more days to find satisfactory values.” [PD]
Power Tuning without PADA:Key Findings
Inconsistency of daily behavior
Nontrivial burden
Limited result
0
20
40
60
80
100
0 0.25 0.5 0.75 1
expe
cted
bat
tery
life
(h)
F1 score
PA PB PC PD PE
Due to own intuition and limited trials
Conclusion
• Challenge of power-aware development for mobile sensing applications
• Repetitive power evaluation in real-life situations
• PADA: Power-aware development for mobile sensing applications
• Context-driven record-and-replay
• We show the usability of PADA with two developer studies
