A Lightweight Computation Offloading System for Mobile Cloud Computing

Abstract

The latest advancements in mobile computing technology have changed human preferences for computing. Mobile devices are increasingly becoming essential and most effective computational tools. However, mobile devices are resource limited and those limitations prevent them from performing complicated tasks and running computation-intensive mobile applications. To overcome those limitations the computation offloading mechanism has been widely considered. The existing computation offloading systems perform several tasks in the mobile side, including application profiling, application partitioning, obtaining and estimation of decision parameters, and offloading decision. Running those components (tasks) in the mobile side brings an extra computation overhead to the resource constrained mobile devices, which affects the performance gain from computation offloading. Moreover, computation offloading requires partitioning of the application into client and server parts for local and remote execution, respectively. A significant effort has been devoted to determine optimal partitioning techniques and to reduce the total execution time of the applications. However, most of the existing computation offloading systems haven‟t considered the intercommunication between the application parts during execution and the result of other solutions remain unsatisfying. To alleviate those problems a lightweight computation offloading system that optimizes the performance of computation-intensive mobile applications is proposed in this research. The components of the proposed system that require heavy computation are implemented in the server side and a lighter offloading engine is designed to reduce overhead in the mobile device. In partitioning, the proposed system considers the call relationship between each method and the execution time of the methods. Instead of evaluating an individual method for offloading, it determines the offloading and integrating points for the whole application using the critical path method. Furthermore, the evaluation results of the proposed computation offloading system showed that it can speedup mobile applications 5.81X in average.