Resource Optimization In Mobile Edge Computing

With the improvement of the reliability and load capacity of wireless networks,and the advancement of the manufacturing process of electronic components,mobile devices will support more intelligent services.Among them,ultra reliable and low latency services bring challenges to the local computing capabilities of the mobile devices,while mobile edge computing is a promising technology to solve the dilemma.In a mobile edge computing network,computing access points provide storage and computing capabilities by connecting mobile devices and central cloud servers.This paper uses traditional algorithms and machine learning methods to jointly optimize time delay and energy consumption delay,and design task offloading strategies.Firstly,this paper focuses on a mobile edge computing network where tasks generated form mobile devices are offloaded to multiple computing access points.Considering that each task can be selected to be offloaded to the computing access points for computing or to be computed locally,the delay includes transmission delay and computation delay,and energy loss includes transmission loss and computation loss.The integer optimization problem is proposed to be solved by the relaxation recovery algorithm based on linear programming and the branch and bound algorithm.In the relaxation recovery algorithm,the integer constraint is relaxed to the continuous solution space,and the optimal solution is obtained through the linear programming tool.Then,the optimal offloading policy for each task is selected according to the single connection constraint.In the branch and bound algorithm,the decision tree describes the solution space,where the follow-up branch is obtained via non-integer decision solved from the relaxed problem of each node in the decision tree.It is proved that the auxiliary service of computing access points reduces the task processing delay and energy loss for the mobile devices.The branch and bound method can obtain the approximate optimal solution,which is better than the linear relaxation recovery algorithm,while the computational complexity of linear relaxation recovery algorithm is far less than the branch and bound method.Then a task offloading assignment method based on cross-entropy learning is proposed.The binary offloading policy can be shaped as bernoulli distribution,and then Monte Carlo method generates effective samples based on the probabilistic distribution.The minimum cross-entropy method is to learn the samples and update the characteristic parameters in the Bernoulli distribution.To reduce the redundant samples,an adaptive sampling method is proposed,which divides a single sample into multiple interrelated sub-blocks,and then generates sub-blocks in turn.The proposed algorithm performs well in terms of computational loss,performance,robustness,etc.,and can be expended for denser mobile edge computing networks.Finally,we focus on the scenario where the tasks generated by multiple mobile devices are transmitted to the fixed computing access point through multiple independent channels.Due to equipping the mobile devices with the ability of dividing the task size,continuous optimization variables is introduced in the optimized problem.A data-model driven method based on neural network is proposed to offloading assignment.Discrete variables is utilized to build branch and bound solution space,then neural network is established to learn the pruning process of the decision tree for branch and bound method until all nodes of the decision tree are searched or pruned.Compared with the traditional branch and bound method,the computational complexity of the proposed algorithm is highly reduced,and the performance of the algorithm is close to the traditional one.