Research On Delay-Sensitive Task Offloading In Distributed Edge Computing

With the continuous improvement of user service requirements,a large number of computing intensive and delay sensitive applications came into being.In order to meet the needs of users,the number of communication devices in wireless networks is increasing,and the amount of communication data is increasing explosively.In order to alleviate the contradiction between the capabilities of end equipment and the requirements of large-scale tasks and meet the more diverse and higher standard task requirements of end users,distributed edge computing appeared.Distributed edge computing saves the resources of the end equipment and improves the execution efficiency of tasks by shorten the distance between end user and edge servers and offloading the tasks that end equipment has no power to execute to edge server.However,compared with the centralized cloud computing data center,the geographical location of edge servers is scattered and the equipment resources are limited.Therefore,designing a reasonable task offloading decision and making rational use of the limited resources in the edge computing network are of great significance to optimize the user service requirements of task processing.Due to the complex network environment and diverse task processing performance requirements,unreasonable task offloading decisions are easy to cause network load imbalance and reduce resource utilization efficiency.Task offloading involves computation offloading and resource allocation.Among them,the computation offloading determines the location,moment and mode of task execution.Resource allocation determines the type,moment and quantity of resources allocated to tasks.Thus,task offloading is the result of multi-faceted coupling.Therefore,how to design an efficient task offloading strategy and realize the scheduling and allocation of task offloading among edge servers is an important problem to be solved in distributed mobile edge computing.To solve the above problems,this thesis focuses on the distributed edge computing network and carries out the research on task offloading in distributed edge computing.The main research contents of this thesis are as follows:(1)The thesis solves the problem of delay-sensitive tasks offloading in distributed edge computing system.In a multi-edge server system,edge servers can be divided into multiple collaboration spaces,edge servers can collaborate with each other in the same collaboration space.Several fine-grained tasks are waiting to be executed in the system.Based on the delay of task execution,this thesis formulates a problem to minimize the average delay in edge computing system.This thesis improves the traditional ant colony algorithm,by using the analytic hierarchy process to optimize the initialization of the pheromone matrix,and modifying the pheromone matrix’s update method and the fitness function’s designing.Task offloading decisions include the choice of offloading mode,access point and execution location.Finally,the simulation results show that the proposed algorithm can make full use of the advantages of edge servers and reduce the average task delay on the basis of ensuring the quality of task processing.(2)The thesis solves the joint optimization problem for delay-sensitive task offloading in distributed edge computing system.In a multi-edge server computing system,bandwidth resource allocation and computation offloading of edge system are considered jointly.First,this thesis summarizes the bandwidth resource allocation problem based on coalition game,and transforms the problem to finding a Pareto optimal solution problem for an array,and solves the problem using a multi-objective particle swarm optimization algorithm.Secondly,for the delay-sensitive task offloading decision,this thesis takes minimizing task delay as the goal and solves the tasks offloading decision.The branch and bound method is used to construct the propagation tree to find the optimal case for the task offloading decision.In order to evaluate edge servers in the tree,the thesis builds an evaluation matrix,which is transformed into a set of evaluation indicators for task offloading decisions.The evaluation matrix is designed considering the number of layers of the edge server in the propagation tree,the computing power of the edge server,and the task’s queue length on the edge server.The simulation results show that the proposed algorithm effectively reduces the task processing delay.