|
|
ENQUIRE PROJECT DETAILS BY GENERAL PUBLIC |
| Project Details |
| Funding Scheme : | General Research Fund | ||||||||||||||||||||||||
| Project Number : | 611813 | ||||||||||||||||||||||||
| Project Title(English) : | Diagnosing Energy Inefficiencies in Mobile Applications based on Observable Sensory Data Utilization | ||||||||||||||||||||||||
| Project Title(Chinese) : | 基于傳感器數據使用率的移動應用程式中能源低效利用问题的诊断方法 | ||||||||||||||||||||||||
| Principal Investigator(English) : | Prof Cheung, Shing-chi | ||||||||||||||||||||||||
| Principal Investigator(Chinese) : | |||||||||||||||||||||||||
| Department : | Dept of Computer Science & Engineering | ||||||||||||||||||||||||
| Institution : | The Hong Kong University of Science and Technology | ||||||||||||||||||||||||
| E-mail Address : | scc@cse.ust.hk | ||||||||||||||||||||||||
| Tel : | 2358 7016 | ||||||||||||||||||||||||
| Co - Investigator(s) : |
|
||||||||||||||||||||||||
| Panel : | Engineering | ||||||||||||||||||||||||
| Subject Area : | Computing Science & Information Technology | ||||||||||||||||||||||||
| Exercise Year : | 2013 / 14 | ||||||||||||||||||||||||
| Fund Approved : | 500,000 | ||||||||||||||||||||||||
| Project Status : | Completed | ||||||||||||||||||||||||
| Completion Date : | 30-6-2017 | ||||||||||||||||||||||||
| Project Objectives : |
|
||||||||||||||||||||||||
| Abstract as per original application (English/Chinese): |
|
||||||||||||||||||||||||
| Realisation of objectives: | Smartphone applications’ energy efficiency is vital, but many Android applications suffer from serious energy inefficiency. Locating these energy problems is labor-intensive and automated diagnosis is thus desirable. This motivates us to tabulate common energy inefficiency issues and developed a framework to analyze energy inefficiency of Android apps in this project. To achieve the first objective, we collected 608 candidates from Google Play Store [17] using a web crawler. These applications have release logs containing at least one of the following keywords: battery, energy, efficiency, consumption, power, and drain. We then performed manual examination to ensure that these applications indeed were infected with energy problems and developers have fixed these problems in the applications’ latest versions (note that we did not have access to their earlier versions containing energy problems). To find interesting open-source subjects, we first randomly selected 250 candidates GitHub and SourceForge. Since we are interested in applications with a certain level of development maturity, we refined our selection by retaining those applications that: (1) have at least 1,000 downloads, (2) have a public bug tracking system, and (3) have multiple versions. We studied our collected subjects’ real bug reports, commit logs, bug-fixing patches and release logs concerning energy problems occurring to open-source and commercial Android apps. Major causes leading to energy inefficiencies were identified. With these findings, we carried out studies for the second objective and developed a framework to automatically diagnosing such energy inefficiencies. A key challenge in diagnosing energy inefficiencies at mobile applications is that the absence of a decidable criterion that facilitates effective assessment on the occurrences of such inefficiencies. We developed a technique to address this challenge. Our approach explores an application’s state space by systematically executing the application using Java PathFinder (JPF). It monitors sensor and wake lock operations to detect missing deactivation of sensors and wake locks. It also tracks the transformation and usage of sensory data and judges whether they are effectively used by the application using our state-sensitive data utilization metric. By doing so, we can generate detailed reports with actionable in-formation to assist developers in validating detected energy inefficiencies. Another key challenge is that Android applications are event-driven and highly hinge on user interactions. Their program code comprises many loosely coupled event handlers, among which no explicit control flow is specified. At runtime, these event handlers are called by the An-droid framework, which builds on hundreds of native library classes. Automated diagnosis requires: (1) generation of valid user interaction events, and (2) correct scheduling of event handlers. To address this challenge, we developed technique to analyze an Android application’s GUI layout configuration files, and systematically enumerate all possible user interaction event sequences with a bounded length at runtime. We showed that such a bounded length does not impair the effectiveness of our analysis, but instead helps quickly explore different application states and identify energy problems. We developed an application execution model derived from Android specifications. This model captures application-generic temporal rules that specify calling relationships between event handlers. With this model, we are able to ensure an Android application to be exercised with correct control flows, rather than being randomly scheduled on its event handlers. To achieve the third objective, a tool called GreenDroid was built on top of Java PathFinder (JPF), a model checker for Java programs. All three proposed project objectives were successfully completed with a set of evaluation datasets and tools implementing the proposed framework made available to the public at: - GreenDroid: The major framework that automatically detects energy inefficiencies at Android apps. The tool and evaluation dataset are available at http://sccpu2.cse.ust.hk/greendroid/. - Elite: A framework component that supplements GreenDroid to automatically detect and locate wakelocks that induce both functional and non-functional issues in Android apps. The tool and its evaluation dataset are available at http://sccpu2.cse.ust.hk/elite/toc.html. | ||||||||||||||||||||||||
| Summary of objectives addressed: |
|
||||||||||||||||||||||||
| Research Outcome | |||||||||||||||||||||||||
| Major findings and research outcome: | The findings of this project were disseminated at five reputable journal articles (2xTSE, IEEE Software, IJSI, SCIS) and seven reputable conference publications (2xICSE, 2xASE, POPL, FSE and ISSTA). Two PhD students were trained. Our publication at ASE 2016 was were highly recognized by the software engineering community and received the ACM SIGSOFT Distinguished Paper award. We summarize selected findings as follows. - Although root causes of energy problems can vary with different applications, majority of them (over 60%) are closely related to two types of coding phenomena. The first phenomenon arises from missing sensor or wake lock deactivation. To use a smartphone sensor, an application needs to register a listener with the Android OS. The listener should be unregistered when the concerned sensor is no longer needed. Similarly, an application needs to acquire a wake lock from the Android OS. The acquired wake lock should be released as soon as the computation completes. The second phenomenon arises from sensory data underutilization. Smartphone sensors probe their environments and collect sensory data. These data are obtained at the cost of battery energy and therefore should be well utilized by applications. Poor sensory data utilization could result in energy waste. - Energy problems are serious. They have covered many types of Android applications and affected potentially millions of users. - It is relatively more difficult to diagnose and fix energy problems, as compared to non-energy bugs; user interaction con-texts and debugging logs can help problem diagnosis, but they require additional reporting efforts, which may not be desirable. - Two common patterns of energy inefficiencies can be identified: missing sensor or wake lock deactivation and sensory data underutilization. - Many energy-efficient mobile apps are written to adjust their behavior dynamically in response to changes in their runtime contexts (e.g., environmental conditions). We found that certain some context changes can result in unstable inconsistency detection of these changes. Such unstable detection can be effectively suppressed by pre-defined change patterns. - Over 27.2% of Android apps utilizes wake locks to protect critical computation. Computational tasks protected by wake locks are often run asynchronously and could be long running, and can bring users observable or perceptible benefits. - Eight patterns are found to commonly accountable for wake lock misuses: unnecessary wakeup, wake lock leakage, premature lock releasing, multiple lock acquisitions, inappropriate lock type, erroneous timeout setting, inappropriate flags, and permission errors. | ||||||||||||||||||||||||
| Potential for further development of the research and the proposed course of action: |
We hope that in future more efforts from research communities and industries can be devoted to designing useful techniques to help developers combat energy and performance bugs in their smartphone applications. We believe that energy and performance diagnosis is becoming increasingly important as mobile applications continually integrate themselves into people’s daily lives. Our findings lead to the following future work. - Extend energy inefficiency diagnosis to other non-functional issues in mobile applications. - Extend energy inefficiency diagnosis to hybrid mobile applications such as those running on top of an Android Webview component. - Automatically locate the root cause of a detected energy inefficiency and repair it accordingly. - Further enrich our current dataset with new issues found in mobile apps launched over emerging Android OS. - Extend the framework to conduct diagnosis for new Android OS. - Adapt the framework to iOS applications. | ||||||||||||||||||||||||
| Layman's Summary of Completion Report: | The mobile app market is expanding rapidly. Many mobile applications are realized in an energy-inefficient or performance-ill way, seriously affecting user experience. There are a lack of powerful tools to combat energy inefficiency problems. In this project, we developed an effective framework for automated energy inefficiency diagnosis in mobile applications. We studied the characteristics of these bugs and developed automated techniques to address three major challenges: (a) how to trigger an energy bug, how to judge the occurrence of energy inefficiency, and how to determine the amount of diagnosis. We released our framework and experimental datasets to the public, and disseminate our research results in reputed research venues. Our results were well received and award-winning. Two PhD students were trained. | ||||||||||||||||||||||||
| Research Output | |||||||||||||||||||||||||
| Peer-reviewed journal publication(s) arising directly from this research project : (* denotes the corresponding author) |
|
||||||||||||||||||||||||
| Recognized international conference(s) in which paper(s) related to this research project was/were delivered : |
|
||||||||||||||||||||||||
| Other impact (e.g. award of patents or prizes, collaboration with other research institutions, technology transfer, etc.): |
|||||||||||||||||||||||||
| SCREEN ID: SCRRM00542 |