검색 상세

Power Modeling, Analysis, and Optimization for Mobile Devices

초록/요약

스마트폰과 스마트시계의 등장으로 모바일 기기들은 점점 더 대중화 되고 있다. 기존의 피처폰과 다르게 스마트폰들은 고성능 하드웨어와 복잡한 소프트웨어로 구성되어 많은 에너지를 소비한다. 더 나아가 스마트시계에서는 제한적인 크기와 무게로 인해서 배터리 자원은 매우 소중히 여겨진다. 따라서 모바일 기기들의 유용성은 증가하는 성능의 요구와 배터리 용량 사이의 관계에 의존되어 있다. 그렇지만 배터리 기술의 증가 속도는 실리콘 업계의 무어의 법칙을 따르지 않고 있다. 배터리 기술 성능의 두 배의 발전은 25년 주기와 같다. 이러한 이유 때문에 배터리 이용성을 향상 시키는 일은 모바일 기기 업계에서 최우선 과제중 하나로 여겨져 왔다. 본 학위 논문에서는 스마트폰과 스마트시계 배터리 문제를 다루는 몇 가지 종합적인 접근법들을 제안한다. 핵심 기여는 개발자와 최종 사용자들에게 모바일 응용프로그램들의 전력 사용 정보를 제공하고 모바일 기기의 배터리를 그들 스스로 관리할 수 있도록 돕는다는 것이다. 첫 번째는 전력 모델과 자동 생성 그리고 추정 프레임워크의 제공을 통해서 개발자들 스스로가 응용프로그램들이 모바일 기기에서 실행 될 때 발생되는 전력 소모를 이해할 수 있도록 한 것이다. 제안하는 프레임워크의 유용성을 보이기 위해서 실제 하드웨어 사용을 수집 했을 뿐만 아니라 여러 스마트폰들의 전력 모델들을 높은 정확도로 생성했다. 두 번째는 비정상적인 종료와 많은 에너지 소모를 발생 시키는 저품질 모바일 응용프로그램들을 제거하기 위해서 소스코드와 사람의 개입 없이도 응용프로그램 상태를 효과적으로 탐색하고 크래시 유무와 과도한 자원 사용 그리고 서로 다른 크기의 화면에서의 호환성 문제를 검증하는 자동 품질 분석기를 제안한 것이다. 이 분석기는 모바일 응용프로그램 마켓들의 품질 유지를 위해서 마켓 관리자에 의해서 사용될 것으로 예상 된다. 마지막으로는 알림 전달에 의한 전력 소모를 최적화할 뿐만 아니라 중복 알림에 의한 정보 과부화를 완화하기 위해서 지능적인 알림 관리 시스템을 제안한다. 불필요한 알림 전송을 예측하기 위해서 제안하는 방법은 실제 데이터로부터 얻어진 알림 메시지들과 상황정보들을 학습하여 개인화 모델을 생성한다. 생성한 기계학습 모델을 이용해서 사용자가 집중하고 있는 적절한 장치에만 알림 메시지 전송을 발생시킨다.

more

초록/요약

Mobile devices such as a smartphone and a smartwatch have increasingly become pervasive in modern life. Unlike feature phones, configured with more powerful hardware and more complex software, smartphones consume much more energy. Moreover, in the smartwatch, the battery resource is very precious due to limited size and weight. Therefore, the utility of mobile devices is dependent on the tension between the increasing demand for their performance and their battery life. Unfortunately, the improvement pace of battery technology does not track Moore’s Law in the silicon industry. Battery technology is equal to a doubling of performance on a 25-year cycle. For that reason, improving battery utilization has become one of the foremost challenges in the mobile device industry. In this dissertation, we propose a few comprehensive approaches to address battery concerns of the smartphone and the smartwatch. Our major contribution is to provide developers and end-users with power usage of mobile applications and help them manage their mobile devices’ battery themselves. Our research consists of three major steps as follows. Firstly, we enable developers to understand power consumption while running their applications on mobile devices by providing power model and its automated build and estimation framework. To demonstrate usefulness of the proposed framework, we not only collect real hardware usage, but also construct several power models of smartphones with high accuracy. Secondly, to filter out low quality mobile applications, having abnormal exit and large energy consumption, we propose an automated quality analyzer that efficiently explores application pages and verifies the presence of crashes, excessive resource usage, and compatibility problems in differently sized screens without source code and human involvement. We anticipate that market curators to maintain quality of mobile application markets will use this analyzer. Finally, we propose an intelligent notification management system not only to optimize power consumption by notification delivery, but also to remedy information overload by redundant notifications. To predict unnecessary notification delivery, the proposed notification management builds a personalized model, learning from context data and notification labels in the wild. With machine learning model, we can trigger notification delivery only to an opportune device on which a user focuses.

more

목차

Acknowledgements i
List of Figures vii
List of Tables x
Abstract xi
1 Introduction 1
1.1 Automated Power Model Generation and Estimation . . . . . . 3
1.2 Automatic Quality Analysis for a Large Number of Mobile Applications
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.3 Intelligent Notification Delivery Management . . . . . . . . . . . 5
1.4 Dissertation Outline . . . . . . . . . . . . . . . . . . . . . . . . 5
2 Automated Power Model Generation Method for Smartphones 6
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.2 Power Model Generation . . . . . . . . . . . . . . . . . . . . . . 9
2.2.1 Step 1: Generation of Chunks . . . . . . . . . . . . . . . 11
2.2.2 Step 2: Filtering Out Invalid Chunks . . . . . . . . . . . 12
2.2.3 Step 3: Power Model Computation . . . . . . . . . . . . 14
2.3 Artificial Usage Pattern Generation . . . . . . . . . . . . . . . . 17
2.4 Evaluation of Power Model . . . . . . . . . . . . . . . . . . . . . 19
2.5 Power Modeling Tools . . . . . . . . . . . . . . . . . . . . . . . 23
2.6 Related Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.7 Discussion and Limitations . . . . . . . . . . . . . . . . . . . . . 25
2.8 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3 QDroid: Automatic Quality Analysis for a Large Number of
Mobile Applications 28
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.2 Motivation and Design Requirement . . . . . . . . . . . . . . . . 32
3.3 QDroid Architecture . . . . . . . . . . . . . . . . . . . . . . . . 34
3.4 Dynamic Exploration . . . . . . . . . . . . . . . . . . . . . . . . 35
3.5 Verifying Multiple Quality Factors . . . . . . . . . . . . . . . . . 37
3.5.1 Crash . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.5.2 Resource Usage . . . . . . . . . . . . . . . . . . . . . . . 38
3.5.3 Compatibility Problem in Differently Sized Screens . . . 38
3.6 Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.6.1 Dynamic Exploration . . . . . . . . . . . . . . . . . . . . 40
3.6.2 Performance Counter . . . . . . . . . . . . . . . . . . . . 40
3.6.3 Modified Android . . . . . . . . . . . . . . . . . . . . . . 41
3.6.4 Logger . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
3.6.5 Summarizer . . . . . . . . . . . . . . . . . . . . . . . . . 42
3.7 Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.7.1 Methodology . . . . . . . . . . . . . . . . . . . . . . . . 43
3.7.2 Activity Coverage and Discovering App Crashes . . . . . 44
3.7.3 Coverage Result . . . . . . . . . . . . . . . . . . . . . . . 45
3.7.4 Resolution Problem . . . . . . . . . . . . . . . . . . . . . 47
3.7.5 Resource Usage . . . . . . . . . . . . . . . . . . . . . . . 48
3.8 Related Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
3.8.1 Mobile Application Page Exploration . . . . . . . . . . . 52
3.8.2 Mobile Application Verification . . . . . . . . . . . . . . 53
3.9 Discussion and Limitations . . . . . . . . . . . . . . . . . . . . . 55
3.10 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
4 PASS: Reducing Redundant Interactions between a Smartphone
and a Smartwatch for Energy Saving 57
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
4.2 Background and Motivation . . . . . . . . . . . . . . . . . . . . 61
4.2.1 Notification Delivery Mechanism . . . . . . . . . . . . . 62
4.2.2 Preliminary Experiment . . . . . . . . . . . . . . . . . . 63
4.3 Auto-Labeling of Notifications . . . . . . . . . . . . . . . . . . . 69
4.4 Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
4.5 Dataset . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
4.5.1 Unnecessary Notification Delivery . . . . . . . . . . . . . 76
4.6 Prediction for an Reply Notification . . . . . . . . . . . . . . . . 77
4.6.1 Building Prediction Model . . . . . . . . . . . . . . . . . 77
4.6.2 Feature Importance . . . . . . . . . . . . . . . . . . . . . 80
4.6.3 Neural Network Models . . . . . . . . . . . . . . . . . . 83
4.7 Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
4.7.1 Prediction Results . . . . . . . . . . . . . . . . . . . . . 88
4.7.2 Model Comparison . . . . . . . . . . . . . . . . . . . . . 90
4.7.3 Energy Saving Effectiveness . . . . . . . . . . . . . . . . 91
4.7.4 Sensing Overhead . . . . . . . . . . . . . . . . . . . . . . 97
4.8 Discussion and Limitations . . . . . . . . . . . . . . . . . . . . . 97
4.9 Related Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
4.9.1 Interruptibility Management System . . . . . . . . . . . 100
4.9.2 Energy Optimization . . . . . . . . . . . . . . . . . . . . 101
4.10 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
5 Conclusion 103
Bibliography 105
Acronym and Abbreviation 120
Abstract in Korean 121

more
반출 Meta View 목록