•   When: Wednesday, November 12, 2014 from 10:00 AM to 12:00 PM
•   Speakers: Mohammed Hassan
•   Location: ENGR 4801
•   Export to iCal

Abstract

The increasing popularity of mobile devices calls for effective execution of mobile applications. Due to the relatively slower CPU speed and smaller memory size, plenty of research has been conducted on properly splitting and offloading computing intensive tasks in mobile applications to external resources (e.g., public clouds). Prior research on mobile computation offloading has mainly focused on how to offload. Moreover, existing solutions require the application developers to specify the computation intensive segment of the applications beforehand. In addition, these solutions do not work transparently with existing applications.

In this dissertation, we aim to design and implement a mobile application offloading framework by addressing these issues. For this purpose, we first design and implement a transparent offloading mechanism called FaST: a Framework for Smart and Transparent mobile computation offloading, to allow mobile applications to automatically offload resource-intensive methods to more powerful computing resources without requiring any special compilation or modification to the applications' source code or binary. We also find the key features that can dynamically impact the response time and the energy consumption of the offloaded methods and thus design Lamp, a runtime Learning model for application profiler, based on which we profile the local on-device and the remote on-server executions of the applications. To address the problem of what to offload, we design and implement an application partitioner, called Elicit, to Efficiently identify computation-intensive tasks in mobile applications for offloading in a dynamically changing environment by estimating the applications' on-device and on-server performance with Lamp.

In addition to the offloading demand for mobile applications, the increasing popularity of mobile devices has also caused the demand surge of pervasive and quick accessing files across different personal devices owned by a user. Most existing solutions, such as DropBox and SkyDrive, rely on centralized infrastructure (e.g., cloud storage) to synchronize files across different devices. Therefore, these solutions come with potential risks of user privacy and data secrecy. In addition, the consistency, synchronization, and data location policies of these solutions are not suitable for resource-constrained mobile devices. Furthermore, these solutions are not transparent to the mobile applications. Therefore, in this dissertation, we design and implement a system Virtually Unify Personal Storage (vUPS) for fast and pervasive accesses of personal data across different devices.