CAMDroid CAM Droid An An Ad Adap aptation tation Fr Fram - - PowerPoint PPT Presentation

camdroid cam droid
SMART_READER_LITE
LIVE PREVIEW

CAMDroid CAM Droid An An Ad Adap aptation tation Fr Fram - - PowerPoint PPT Presentation

HumanSys 2017 CAMDroid CAM Droid An An Ad Adap aptation tation Fr Fram amewo ework rk fo for And r Andro roid id Cont Co ntext ext-Awa Aware re Mu Multitasking ltitasking Kouemo Ngayo Anatoli Dimitrov 1 , Xi Xiao aolon


slide-1
SLIDE 1

CAM CAMDroid Droid

An An Ad Adap aptation tation Fr Fram amewo ework rk fo for And r Andro roid id Co Cont ntext ext-Awa Aware re Mu Multitasking ltitasking

Kouemo Ngayo Anatoli Dimitrov1, Xi Xiao aolon

  • ng Zh

Zhen eng1, Fu Xiao2

1Tsinghua University 2Nanjing University of Posts and Telecommunications

P.R. China

HumanSys 2017

slide-2
SLIDE 2

Mu Multit ltitasking asking

2

  • Multitasking
  • Perform multiple tasks (also known as processes) over

a certain period of time by executing them concurrently.

  • Android supports multitasking
  • Starting from Android 4 in 2013

Is it satisfactory?

slide-3
SLIDE 3

An Android droid Mu Multit ltitasking asking

3

  • Foreground
  • State: Running
  • Active and interactive
  • Background
  • State: Sleeping/ Closed
  • Suspended to save energy
  • Only interact with foreground App
  • Due to one small screen
  • not executing concurrently!

In current multitasking, Apps are sleeping instead of concurrently running in background!

slide-4
SLIDE 4

Re Research search Ta Target rget

4

  • Context-aware multitasking
  • Apps run in the background
  • “Real” concurrent execution
  • Enable users interact with background Apps
  • Dynamically preload/offload Apps to reduce the launch

time/save the memory resource.

slide-5
SLIDE 5

Ch Chall allenges enges

5

  • Background Apps are suspended and cannot

access whole context information

  • Keep all Apps running in background will lead to

unacceptable energy consumption

  • Use up the memory
slide-6
SLIDE 6

Co Context ntext Awa Awareness reness

6

  • Sense and react based on the physical conditions
  • Context types:
  • location, identity, activity, time etc.
slide-7
SLIDE 7

Co Context ntext Awa Awareness reness

7

  • Widely used in mobile Apps
  • Example: Location-based preloading

Apps maintain the context by themselves Isolated adaption engine is used in own App

slide-8
SLIDE 8

CAr CAreDroid eDroid for for Co Context ntext-aware aware Ap Apps ps

8

  • External context (outside OS) only
  • Without internal context (App status inside OS)
  • Foreground bias
  • Interact with foreground Apps only
  • Static configuration written by App developers
  • Preform predefined actions in the given context

[1] Salma Elmalaki et.al, CAreDroid: Adaptation Framework for Android Context-Aware Applications, MobiCom 2015.

Context-aware multitasking demands dynamic control of background Apps based on both external and internal context

slide-9
SLIDE 9

CAM CAMDroid Droid

9

  • Context-aware multitasking
  • Dynamic control of background Apps
  • With both external and internal context

Adaptation Engine Context Analyzer External Context Internal Context Foreground App Background Apps

slide-10
SLIDE 10

Ou Our r So Solutions lutions

10

  • Background Apps are suspended and cannot

access whole context information

  • Context analyzer inside OS to collect both external

and internal context for all Apps

  • Keep all Apps running in background will lead to

unacceptable energy consumption

  • Adaptation engine that preloads or executes Apps that

are frequently used in recent period, in current context

  • use up the memory
  • Activate Apps with strict memory constraints
slide-11
SLIDE 11

CAM CAMDroid Droid

11

  • Context-aware multitasking
  • Dynamic control of background Apps
  • With both external and internal context

Adaptation Engine Context Analyzer External Context Internal Context Foreground App Background Apps

slide-12
SLIDE 12

Co Context ntext An Analyzer alyzer

12

  • External context
  • Analyze with sensor and sensorless sensing
slide-13
SLIDE 13

Co Context ntext An Analyzer alyzer

13

  • Internal context
  • App status, number of use, service time, required

memory size …

  • Hook system calls
slide-14
SLIDE 14

CAM CAMDroid Droid

14

  • Context-aware multitasking
  • Dynamic control of background Apps
  • With both external and internal context

Adaptation Engine Context Analyzer External Context Internal Context Foreground App Background Apps

slide-15
SLIDE 15

Ad Adaptation aptation En Engine gine

15

  • Real-time multitasking with context-awareness
  • Foreground/background Apps react accordingly
  • Preload/offload apps
  • Current implementation
  • most frequently used in recent period

Memory constraint

𝑤𝑗 = 𝜕1𝑢𝑗 + 𝜕2ℎ𝑗 + 𝜕3𝑑𝑗

Metric App i will be used Dynamic updating the score

slide-16
SLIDE 16

CAM CAMDroid Droid Im Implementation plementation

16

  • Device & Operating System
  • Android 5.1.1
  • Google LG Nexus 5 mobile phone
  • System image size
  • Android:

358930 KB

  • CAMDroid:

380851 KB

  • Overhead:

21921 KB

slide-17
SLIDE 17

Ev Evaluation aluation

17

  • Predicted task list
  • If the opened App is in the list, we regard CAMDroid

accurately predicts once.

  • 100 trails under different external contexts
  • Our tested contexts are coarse-grained

Others At home Running At work

slide-18
SLIDE 18

Ev Evaluation aluation

18

  • Reduce the launch time
  • Due to the preloading, launch time is reduced
  • Reduced by 50% in average
slide-19
SLIDE 19

Ev Evaluation aluation

19

  • Off-loading saves energy
  • Close Apps unlikely used in current context

5 min

Event Tracker CAMDroid Android Battery level drops 4% in native Android, and 3% in CAMDroid, during 30 minutes

slide-20
SLIDE 20

De Demo mo

20

slide-21
SLIDE 21

Co Conclusion nclusion

21

  • CAMDroid -- Context-Aware Multitasking
  • Bring context-awareness into the operating system
  • Provide external and internal context to Apps
  • Enable the interaction between user/environment and

background Apps

  • Save energy and launch time
  • Future work
  • Improve prediction accuracy according to fine-grained

correlation between context and App

  • Include personalized models
slide-22
SLIDE 22

Th Than ank k you! you!

Xia iaol

  • lon
  • ng Zheng

Zheng

http://www.greenorbs.org/people/xiaolong/

2017.11.05

HumanSys 2017