Towards Energy Effective Hybrid Execution in Mobile Cloud Computing

Abstract

The Mobile Applications are becoming increasingly pervasive and provide rich functionality on mobile devices. Meanwhile, such devices having strong connectivity with more powerful commercial clouds. The advances of these Cloud Computing and Mobile Computing technologies enable the Mobile Cloud Computing paradigm. Three approaches have proposed as mobile cloud ap- plications. The rst approach enables the mobile devices to access cloud ser- vices. The second approach enables the mobile devices to work collaboratively in ad-hoc manner as cloud resource providers. The nal approach enables the mobile devices to augment the execution of mobile applications by using cloud resources. We have done an extensive study on augmenting execution of mobile appli- cations by using cloud resources and propose a hybrid execution framework to optimizing execution of the application such that the application has minimum energy consumption in processing a ow of data. The proposed approach pro- vides run-time support for dynamically partitioning and execution of mobile application. Our solution also allows mobile devices to overcome resource lim- itations such as CPU capability while leveraging the elastic cloud resources. In this dissertation, we have explained an application framework to provide runtime support for the adaptive partitioning of mobile cloud applications. The framework is able to serve large number of mobile devices by leveraging the elastic cloud resources in existing cloud infrastructures. We have also de- signed a genetic algorithm to solve the partitioning problem. Once the control parameters of the genetic algorithm are de ned, it will be able to provides the optimal partitioning solution. iii The evaluation of the proposed hybrid execution framework can be divided into two phases. The rst phase is to evaluate the performance of genetic algorithm for di erent control parameters and the second phase is to evalu- ate overall framework performance according to the inputs parameters of the genetic algorithm. In rst phase, we have evaluated genetic algorithm for di erent control parameters such as population size, maximum generations count, etc. to clarify which ranges of parameter values are most suitable for di erent input parameters to get the maximum e ciency. In second phase, we have used control parameters which we identi ed in rst phase for genetic algorithm and di erent input parameters such as data ow application size, computational intensiveness of the application, etc. to evaluate overall per- formance of the hybrid execution framework. Moreover, we have compared hybrid execution approach with local execution approach and cloud execution approach. The overall evaluation results of the hybrid execution framework shows that hybrid approach can achieve improvement in computational intensive appli- cation performance over the local execution approach and cloud execution approach. Furthermore, we have shown that the genetic algorithm could be ne-tunned to get energy e ciency on di erent kind of mobile cloud applica-tions such as di erent sizes and di erent computational intensivenesses.