Research On Resource Allocation For Mobile Cooperative Cloudlet Computing Systems

Smart mobile devices is becoming an indispensable part of people's life. However, it has never been able to overcome the contradiction between the portability and the limit CPU, storage and battery until Mobile Cloud Computing (MCC) was proposed. In MCC, the mobile application is transferred to the cloud resource and processed by it, so as to achieve the purpose of saving their own resources. The resources utilized in MCC include remote public clouds and nearby cloudlets. Compared with remote cloud, cloudlet has the advantage of shorter communication delays, but it also has the disadvantage of limited computing resources in a single AP. The thesis proposes Mobile Cooperative Cloudlet Computing (M3C) system to improve the efficiency of cloudlet. In M3C system, the mobile device application is modeled as workflow which is composed by tasks and logical constraints between tasks. To shorten the execution time, tasks are allocated to Access Points (APs) and processed there, which can improve the parallelism of workflow processing and shorten the waiting time. This thesis studied the computation resource and communication resource allocation problem for the M3C system.Firstly, the Particle Swarm Optimization (PSO), Hybrid Earliest Finish Time (HEFT) and Partial Critical Path (PCP) based computing resource allocation algorithms are considered. PSO algorithm is a random search algorithm which can achieve better results. However, the number of iterations needed by this algorithm is large. HEFT and PCP algorithms belong to the list heuristic algorithms. The allocation process consists of two parts:the calculation of list and task allocation according to the list order. The differences between them are the strategy generating the list and the strategy used to allocate task. Specifically, the HEFT algorithm will sort tasks in accordance with the priority order and assign the task to the AP which is able to finish it at the earliest time, while the PCP algorithm will classify the tasks into several collections (i.e., PCP) based on task priority and then allocate the tasks in the same collection to the same AP.Secondly, the thesis studied the wireless resource allocation problem for the M3C system. The constraint is that the distances between APs having data to be transferred between them must be less than the threshold. A dynamic programming method is used to solve the problem of how to deploy APs. The positions of APs shall meet the distance constraints and maximize the coverage area. Then, during channel assignment stage, the mutual interference between the links shall be avoided. A channel allocation procedure is proposed which is used to achieve a uniform channel allocation scheme.Finally, a series of simulation experiments are performed to evaluate the performance of M3C, which includes computation resource allocation, AP deployment, and channel resource allocation. The proposed M3C system is compared with the ICloudlet system to prove that cooperation of APs can shorten the process time of workflow. Further, the impact of many parameters on the system performance are studied, includ-ing the number of APs, the number slice server, the amount of computing resource in one AP, the channel assumption, and the amount of communication resource. Simulation results show that the proposed M3C system can utilize the computation and communication resources effectively and the user experience can be improved.