A Dynamic Framework to Support Offloading under Resource Constrained Environment

Abstract

Mobile phones have become an inevitable part in human life. In spite of the rapid growth in mobile technology resource scarcity is still an issue for mobile devices, they lack in computational intensive tasks such as speech recognition, image processing and augmented reality etc. As a solution a pervasive computing technique "Cyber Foraging" can be applied, where mobile devices with limited resources can offload their computational tasks or heavy work to stronger surrogate machines in vicinity. Offloading is considered as an effective method to decrease source consumption of mobiles. It extends the mobile life and save battery. In cloud computing, Cloudlet is an emerging technique to scale up the computational abilities of mobile devices. It allows mobile devices to communicate with resource rich available servers in their local network and mobile devices can offload their computational tasks to these high performance servers and results can be send back to mobile in real time. Our study demonstrate that high-powered cloudlets support various delay sensitive applications that can require computation in real-time. Contrary to previous works and proposed offloading techniques, that only focus specific offloading issues. Our proposed architecture based on various perspectives like automated selection of cloudlet, based upon available resources, an intelligent time based decision engine to decide between local or cloudlet computation. In this thesis we focus on the integration of mobile and cloudlet technologies to develop a time and energy-aware mobile computational task offloading paradigm. In the proposed solution we took android mobile as a client of cloudlet server. To analyze the performance of offloading we choose face recognition as offloading task, as real time face recognition requires abundant resources and energy. Our proposed framework dynamically identify the cloudlet server in mobile phone vicinity and get the information of resources available on servers. The framework includes a time based decision engine which can intelligently decide between local and remote execution. In this way we, consider the dynamic changes and offloading conditions of the context, since offloading may not be beneficial in all the conditions. In order to benchmark our proposed framework, a face recognition application is implemented locally at mobile node. Moreover, similar application has been accessed through application-as-a-Service module implemented in the proposed framework. 7 We’ve conducted different experiments and analysis which shows that by offloading the intensive computations to the cloudlet we can save a significant amount of energy and mobile resources.