Research On Offloading Decision Algorithms In Mobile Edge Computing System

In recent years,with the development of intelligent mobile terminals,the demands for applications from terminals are also increasing.However,these applications may easily cause excessive processing delay and energy consumption when they are executed at the terminals as tasks because of the limitation of the battery capacities and resources in mobile terminals.Mobile Edge Computing(MEC)system has abundant computing and storage resources and possesses the characteristic of proximity access.It allows mobile terminals to offload tasks to the edge cloud,thereby reducing the processing delays and energy consumptions of tasks.In this thesis,the offloading decision algorithm in MEC system is studied deeply.A cooperative offloading scheme based on Markov Decision Process(MDP)and a fast offloading algorithm based on Support Vector Machine(SVM)are proposed,respectively.The proposed scheme and algorithm are analyzed and validated by computer simulations.The main work of the thesis is as follows.(1)For slow-moving scenarios of terminals,the MDP-based Cooperative Offloading Scheme(MDP-COS)is proposed.The scheme consists of two modules: the task partition module and the MDP-based offloading decision module.According to the samples of task sizes,the task partition module first calculates the threshold of task partition with an iteration algorithm,and then uses a task partition algorithm based on threshold and available resources to partition tasks into multiple sub-tasks.The system's state space,action space and reward function of the MDP model are defined,and the state transition probability is deduced in the MDP-based offloading decision module.The state space takes the dynamic changes of the number of tasks in the task queues,the number of edge clouds and terminals in the Ad Hoc cloud accessible to the terminal,the mobility of the terminal and the quality of the wireless channel into consideration.According to the MDP model and the observed system state,the offloading decision strategies of all sub-tasks are obtained with value iteration algorithm,and the sub-tasks are offloaded to the edge cloud,Ad Hoc cloud and central cloud in parallel for cooperative processing.Finally,the time complexity of the scheme is analyzed,and the performance of the scheme is evaluated by computer simulation.The simulation results show that the proposed scheme can effectively reduce the energy consumption of the terminals and completion delay of the tasks.(2)For fast-moving scenarios of terminals,the SVM-based Fast Offload Algorithm(SVM-FOA)is proposed.The algorithm mainly includes two processes: the off-line training process and the online decision process.In the off-line training process,the training data generation algorithm based on decision tree is used to generate training data,and the SVM-based offloading decision function is obtained through training.In the online decision process,the number of sub-tasks which are partitioned is predicted based on the moving speed of the terminal and the maximum allowed offloading delay,then the task is partitioned with a task partition algorithm based on available resources,and each sub-task is transformed into sub-task vector according to the data size of each sub-task and the observed system characteristics.Then,the offloading decision strategies of the sub-task vectors are obtained according to the decision function,and the sub-tasks are offloaded in sequence to the MEC servers for co-processing,when the terminal moves through them according to the strategies.Finally,the time complexity of the algorithm is analyzed and the performance of the algorithm is evaluated by computer simulation.The simulation results show that the proposed algorithm has a fast offloading decision speed and can effectively reduce the average delay of offloading.