Research On Elastic Partition Approach In Mobile Cloud Computing

In recent years, mobile applications and devices have enjoyed rapid development. Compared with current PC, mobile devices still cannot run data intensive applications, and are restricted by the batter power, screen size and wireless communication, etc. To solve the problem, we can adopt mobile cloud computing, which is defined as the ex-tension of cloud computing. The most attractive features of cloud computing lies on the capability in powerful computing and massive data storage as well as a new busi-ness model, which deliveries the computing resources as a utility. The foundation hardware consists at least partly of mobile devices in mobile cloud computing. This definition recognizes the opportunity to harness collective sensing, storage, and com-putational capabilities of multiple networked wireless devices to create a distributed infrastructure that supports a wealth of new applications. Moving the traditional mobile application into the platform, it will be the elastic partition problem.Elastic partition problem is complicated. Elastic partition results will be influence by many factors, such as inputs, constraints and partition goals. Inputs include battery power, connection quality, device loads, cloud loads and performance, etc. Constraints represent the resource cost model, users’requirements, etc. Partition goals can be the minimize cost, the maximize performance, the maximize robustness and the maximize security. Hence, the partition approach should base on above factors.The elastic application model in according to context-awareness could break through the resource constrains. According to the model, we proposed the applica-tion-based elastic partition cost graph, the architecture, the improved (K+l) coarse par-tition algorithm and the partition cost module. The partition algorithm is used to parti-tion the cost graph, where the vertexes are represented by the execution cost on mobile device and offloading cost to cloud. The two factors are represented by some contextu-al information including execution time, current power, network and the probabilities which are obtained by the statistical analysis of historical results. The levels of con-text-awareness could adjust the weight of the contextual information and lead to parti- tion again. Partition cost module is used to store and compute the contextual infor-mation. The extensive experiments deployed the OCR project on the proposed archi-tecture demonstrate a good performance in different input and network environments.