Cooperative Task Schedulingin Mobile Edge Computing System

With the popularization of mobile devices and the development of mobile communications technology,compute-intensive services propose great challenges to the storage capacity and computing capability of mobile terminals.Many researchers consider solving this problem by introducing the thought of cloud computing to the mobile communication network to improve the user experience.Unfortunately,because of the long distance and complicated physical environment between users and the core network where cloud servers usually deployed,offloading between mobile terminals and cloud servers often involves a significant delay in transmission.In view of this,some researchers put forward the concept of mobile edge computing which provides users computing and processing service nearby by deploying MEC servers at the edge of the mobile core network.The computing and storage capacity of MEC servers are much higher than that of mobile terminals and offloading tasks to MEC servers can achieve a reduced processing delay which largely solves computing resources shortage problems in the traditional mobile cloud computing infrastructure.Meanwhile,mobile terminals can communicate with the MEC server directly through one or two jumps as a result of the MEC server always deploys in the edge of the core network where users are.Thus offloading tasks to nearby MEC server can reduce the transmission delay greatly during the offloading process and reduce the burden of the backhaul link in the core network.Besides,for those tasks that need to be offloaded to MEC system,we consider to co-scheduling the computing resources in terminals,MEC servers and the core cloud server.Under this consideration,both the load of MEC server and the average delay of executing tasks can be reduced.Hence,in this paper,our research works investigate the task co-scheduling strategy in the MEC system.First of all,we study the clustering problem in a SDN controller before task scheduling.Through clustering,tasks with similar delay sensitivity and computing resources consumption can be gathered.For the SDN controller,tasks,whose delay are more sensitive and consume less computation resources,take precedence on execution which can effectively reduce the queuing delay of tasks afterwards.This paper establishes a clustering mathematical model for tasks from mobile terminals and compares the impact of different clustering results acquired by several commonly used clustering strategies on co-scheduling strategies.The simulation results show that use the improved k-means clustering algorithm and divide tasks into five clusters are best for reducing system overhead.Next,based on the MEC system in the cellular mobile communication network,this paper mainly studies two “terminal-MEC server” joint resource allocation strategies.For the first STM(Single Task Mode)strategy,a greedy strategy is used to minimize the cost of a single task.Resources in both mobile terminals and MEC servers are scheduled in real-time for each task dependently.For another MTM(Multiple Task Mode)strategy,we establish a multi-task joint resources scheduling model and the optimal target is to minimize the cost of multi tasks.An LPT rule based algorithm is proposed to solve the multi-task “terminal-MEC server” joint resource allocation problem.Numerical results show that both two strategies proposed in this paper can reduce the cost effectively under the premise of guaranteeing the quality of service.Moreover,it is necessary to study offload tasks from the MEC server to remote centre servers.For the sake of this problem,we establish a mathematical model for the problem of minimizing the total cost of executing tasks in MEC servers.Since this problem is NP hard,we present two resource allocation strategies in this paper.The first strategy is dynamic scheduling-based task scheduling strategy which needs to traverse all the combination of tasks.Another strategy is based on genetic algorithm.By simulating the process of biological evolution,it can obtain a result in a relatively short period of time.Comparing to execute all the subtasks by MEC server and remote centre servers separately,simulation results show that both dynamic scheduling-based strategy and genetic algorithm-based strategy can reduce the cost of completing tasks.Besides,the delay of executing all tasks obtained by the genetic algorithm-based strategies is better than the delay obtained by dynamic scheduling-based strategies.In general,this paper investigates the tasks co-scheduling strategy in MEC system and proposes strategies to reduce the cost effectively under the premise of guaranteeing the quality of service which make effective use of the computing resources and provide new ideas to the tasks scheduling and offloading problems for future research.