Joint Task Offloading And Resource Allocation Algorithm Based On Mobile Edge Computing

The fast development of mobile networks and smart devices drives the emergence of diverse computing-intensive applications,i.e.,cloud gaming,virtual reality and graphics recognition.However,the intensive computing requirements of these emerging applications challenge the task execution capabilities of mobile smart devices.Mobile cloud computing(MCC)technology can meet the above-mentioned challenges to a certain extent by assisting mobile smart devices with remote cloud servers to execute intensive computing applications.However,the remote interaction between mobile smart devices and remote cloud server may lead to long communication and computation latency.To tackle the drawbacks of MCC,mobile edge computing(MEC)deploys MEC servers with certain computing capabilities at the edge of mobile networks,and allows MEC servers to help mobile users execute the offloading task of the users,hence,the task execution energy consumption and latency of smart devices can be reduced significantly,and the quality of service of users could be enhanced efficiently.For MEC system,considering the characteristics of user tasks and the characteristics of system comprehensively,and determining reasonable task offloading and resource allocation strategies can help improve user task execution efficiency.The thesis meanly studies the task offloading and resource allocation algorithms of MEC systems,including the following specific contents:First of all,the basic concepts,key technologies and related application senarios of MEC are elaborated,and the current research on task offloading and resource allocation algorithms of various MEC systems is analyzed and summaried.Secondly,the joint CNN layer scheduling,task offloading and resource allocation optimization problem in a multiple tier cooperative MEC system deployed with convolutional neural network(CNN)is formulated as an overall task latency minimization problem,considering constraints including CNN layer scheduling,task offloading,resource allocation,maximum tolerable task execution latency and minimum transmission rate.Because the above problem belongs to a non-convex mixed integer nonlinear programming(MINLP)problem that cannot be directely solved,this thesis first converts the original optimization problem into three subproblems,i.e.,CNN layer scheduling subproblem,task offloading subproblem and resource allocation subproblem,then solves these subproblems successively with extensive search algorithm,reformulation-linearization-technique(RLT)and Lagrange dual method,respectively,to determine the joint optimization strategies.Thirdly,the joint task offloading and resource allocation problem in a dynamic MEC system scenario where user tasks arrive randomly is formulated as an overall user long-term average power consumption minimization problem,considering constraints including task offloading,power allocation,transmission bandwidth allocation,computation resource allocation,task buffer queue limited queue length and user task average latency.Since the formulated optimization problem is a stochastic dynamic programming problem,which cannot be solved directely and conveniently.This thesis first converts the original problem into a deterministic single time slot optimization problem by applying Lyapunov optimization algorithm,then decomposes the deterministic single-slot optimization problem into two subproblems,i.e.,task offloading subproblem and resource allocation subproblem,and uses Kuhn-Munkres(K-M)algorithm and closed form solution in turn to determine the task offloading and resource allocation strategy.Finally,the main research contents of this thesis are summarized,and and the research directions that can be further expanded are analyzed.