An Adaptive Code/Object Offloading Framework for Personalized Big-Data Computing

• Main Authors: Jeng-Peng Shieh, 謝正鵬
• Other Authors: Shih-Hao Hung
• Format: Others
• Language: en_US
• Published: 2015
• Online Access: http://ndltd.ncl.edu.tw/handle/60606246242372908678

博士 === 國立臺灣大學 === 資訊工程學研究所 === 103 === Smart mobile device and wireless networks are reshaping the way people execute applications and access to information. Thus, more and more people rely on social networks and on-line transaction services. As a result, personal data are spreading everywhere in the Internet. However, many users are sensitive to privacy issues and would like their personal data to be handled like personal assets. At the same time, people also share and exchange personal data privately. It is a dilemma that none of the existing mobile applications and public cloud services can resolve. Even though there is a significant progress of hardware and software technologies in recent year, many mobile applications do not perform well due to the shortage of resources for computation, data storage, network bandwidth, and battery capacity. While such applications can be re-designed with clientserver models to benefit from cloud services, the users are no longer in full control of the application, which has become a serious concern for data security and privacy. To protect the personal data, we propose Virtual Phone as a Service(VPaaS) framework to allow the user to control the deployment and execution of applications in the virtual environment. In addition to offloading workload from a physical environment, the virtual environment presents opportunities to enhance the functionalities of the execution environment in the perspective of performance speedup and isolation of the user-managed environment. Existing Android applications can be efficiently accelerated by the framework without any modification during the whole process. We further propose an automatic application offloading scheme, called MobileFBP, which dynamically takes advantage of the personal application clouds to handle sophisticated workload on-demand based on the profiling information and network metrics. MobileFBP enables the programmers in developing dataflow applications that can be executed in mobile-cloud environments. A typical Android application written in Java can be easily converted to FBP as one large task component initially and further broken down into multiple components by declaring the components and expressing the data flow between the components with the assistance from performance profiling tools. Finally, we have developed a framework, COzone, which integrates the above technologies with open source packages to prove the concept of adaptive object offloading and showcase multiple offloading modes. The user can use a weak device to offload and execute personal big-data applications in a personal virtualized environment with the docker container to have personal data processed safely and privately. Depending on the security requirement, the framework is a viable way to avoid the security concerns of exposing private data to public cloud services. We have conducted a case study with three personal data analytics applications to show that object offloading could effectively augment the computing power on behalf of a weak device and save its battery energy as well. The case studies also illustrate how easy it is to augment the application with our API to enable the object offloading capability in the COzone framework.