Research On Offloading Strategy In Energy-Saving Mobile Edge Computing System

The rapid development of wireless communication technology has promoted the rapid development of mobile Internet,and a huge amount of data is generated all the time on mobile devices at the edge of the network.Compared with cloud servers in the network center,mobile devices have limited computing resources,storage resources,and battery power consumption,making it difficult to process so much data in real time.In order to solve this increasingly prominent problem,the emerging computing framework-Mobile Edge Computing(MEC)came into being.Mobile edge computing has the advantages of low latency,energy saving,and reduced core network congestion.This article is about the advantages of energy saving to study mobile edge computing.This paper proposes three edge computing scenarios,which are single-user-multi-MEC server scenarios,multi-user-single MEC server scenarios,and multi-user-multi-MEC server scenarios.In these three scenarios,the problem of offloading decision and resource allocation is studied separately.The common goal is to make the total energy consumption the lowest under the delay limit of the task.In a single-user-multi-MEC server scenario,there are several computing tasks on the user's mobile device waiting to be offloading.The optimization variables are the offloading decision of the task and the mobile device transmit power allocation.In this problem,the AOA(Alternately Optimizing Algorithm)algorithm is proposed.The first step is to obtain the offloading decision by the KM(Kuhn-Munkras)bipartite graph matching algorithm.The second step is to obtain the transmit power allocation through the mathematical proof.In the multi-user-single MEC server scenario,communication resources(subcarriers)and computing resources(CPU frequency of MEC server)are allocated centrally by the MEC server.Because the optimization variables are discrete integers and continuous decimal,the general optimization method is difficult to solve,so machine learning regression algorithm is adopted.Regression algorithm is used to simulate the non-linear relationship between input data and output data,and the output is continuous.In order to obtain training data,10000 pieces of data are randomly generated and labels are exhausted by discretizing the input space.Then three regression algorithms are compared:random forest,Xgboost and artificial neural network.By calculating the corresponding energy consumption of the system,it is found that the artificial neural network can achieve a resource allocation close to the optimal solution.In the multi-user-multi-MEC server scenario,the optimization variable is the offloading decision of mobile devices.Note that the number of mobile devices is K and the number of MEC servers is M.Then the solution space has K*(M+1)possibilities.Using artificial fish swarm algorithm,the optimization variables are mapped to the position of the artificial fish,and the objective function and restriction conditions are converted to the food concentration function.By discretizing the step size and limiting the upper and lower bounds of coordinates,the solution is iterated.Because the algorithm can not guarantee to jump out of the local minimum,the solution obtained is suboptimal.Compared with random unloading mode and non-unloading mode,it is found that lower system energy consumption can be achieved.