Research On Distributed Optimization-based Task Scheduling And Device Collaboration Algorithms In Mobile Edge Computing

In recent years,with the development of mobile Internet and Internet of Things,computation intensive applications and delay-sensitive applications have emerged.Obviously,it is impractical to run such applications on mobile devices.In previous studies,users offloaded computing tasks to the cloud to meet the high computational demands of the device.However,cloud-based solutions result in huge extra bandwidth usage and unpredictable long latency.To solve this problem,mobile edge computing deploys servers at the edge of the network,providing close-range,low-latency services.With the increase of data size,the computing mode of a single mobile edge server will bring tremendous computing pressure.To solve this problem,we propose calculate in a coordinated way.On the one hand,it considers how to encourage the collaborative computing of the device when the device is selfish;on the other hand,when the device is willing to cooperate,how to conduct collaborative computing,such as how to calculate when there are problems such as scalability and privacy in the machine learning task research.In order to solve the above challenges,we propose to use the mobile edge server as the center and utilize the surrounding available resources to further improve the computing performance of the mobile edge computing system.The problem is solved by the distributed optimization method of alternating direction multiplier method(ADMM).The main contents and innovations are as follows:Firstly,to deal with the problem of task scheduling among multi-edge computing devices of a single mobile edge server,we propose a distributed task scheduling algorithm.The mobile edge server can improve the computational efficiency by summarizing the available resources near the edge.Current research only focuses on the scenario of solidarity and collaboration,we further consider the selfishness of edge computing devices.We believe that edge computing devices are rational and usually do not collaborate with mobile edge servers.Therefore,it is necessary to design the effective incentive mechanism to encourage the collaboration of edge computing devices.To solve this problem,we propose the Stackelberg game algorithm based on ADMM.By transforming the non-convex problem to a convex one,the task scheduling problem is solved by using the Stackelberg game theory and the distributed optimization technology ADMM.The convergence,stability and scalability of the proposed algorithm are verified by experiments.Secondly,in the scenario of collaborative computing between single mobile edge server and multiple Internet of Things devices,we believe that the Internet of Things devices have the possibility to perform tasks together.We consider the regression problem in machine learning.Centralized solutions based on mobile edge servers may lead to scalability,computing performance and privacy issues.In order to meet these challenges,we propose a distributed device coordination algorithm based on ADMM,which decomposes the problem into a set of sub-problems by introducing auxiliary variables.Due to the non-differentiability and complexity of the sub-problem,the sub-differential algorithm and improved conjugate gradient method are designed for the sub-problem.Experimental studies on typical data sets show that our algorithm can quickly converge to the optimal solution of the centralized method.Compared with traditional centralized method and independent method,our algorithm performs well in terms of scalability and efficiency.