Secure mobile application partitioning in energy-efficient cloud computing

Abstract

Many emerging mobile applications nowadays tend to be computation-intensive due to the increasing popularity and convenience of smartphones. Nevertheless, a major obstacle prohibits the direct adoption of such applications and that is battery lifetime. Mobile Cloud Computing (MCC) is a promising solution that suggests the partial processing of applications on the cloud to minimize the overall power consumption at the mobile device. However, this does not necessarily save energy if there is no systematic mechanism for evaluating the effect of offloading the application into the cloud. In this research, we develop a mathematical model that represents this energy consumption optimization problem and extend a minimum-cut based algorithm to dynamically solve it. Furthermore, we propose a proprietary protocol - Free Sequence Protocol (FSP) - that allows the dynamic execution of any application according to its call-graph. Since offloading exposes data to eavesdropping attacks, we extend the model to account for security operations including encryption-decryption and conditional offloading. An existing Android image processing application -ImageProx- was altered according to the FSP protocol and integrated into a complete system that involves a profiler and decision engine. The profiler uses mathematical models developed by intensive experiments on the application itself in addition to Wi-Fi and 3G networks. The power measurements were done using PowerTutor application which reaches a 95percent accuracy of the correct values. Our experimental setup consisted of HTC Nexus One - the official device on which PowerTutor was implemented - and a Java server deployed on Amazon EC2 to act as the cloud. The results demonstrate the effect of workload amount, network type, computation cost of functions, and security operations on the overall power consumption. Moreover, they prove that dynamic partitioning successfully provides energy savings at the mobile phone and supersedes the performance of both permanent local and remote execut