Research On Computing Resource Deployment And Task Scheduling For Edge DataCenter Optical Networks

With the rapid development of edge computing,edge datacenter as an infrastructure to provide storage and computing capabilities has been widely concerned,while optical network plays an important role in the Internet of edge datacenter with its low delay and large capacity transmission capability.Edge datacenter optical network,which integrates edge computing technology and optical network technology,has become one of the important application scenarios of the future network.Edge datacenter optical network presents two development trends:"edge follows network" and "dense edge interconnection".Focusing on the above trends,there are problems of computing resource deployment and task scheduling with multiple related sub-tasks.In view of the above problems,this paper focuses on edge datacenter deployment,coflow allocation and geo-distributed data processing,forming the following three research contributions.(1)A hierarchical deployment and load distribution mechanism of edge datacenter for differentiated demand is proposed.Aiming at the problem of computing resource deployment under the trend of "edge following network",the hierarchical deployment and load distribution strategies of the edge datacenter are proposed.Firstly,by analyzing the characteristics of delay,bandwidth and deployment cost at the nodes of distributed processing unit(DU)and centralized processing unit(CU)in 5G wireless access network(RAN)architecture,the cost model,network delay model and processing delay model are established.Secondly,considering the limitation of computing and network resources,the problem of edge datacenter deployment and load distribution is described as a mixed integer nonlinear programming(MINLP)problem with the goal of minimizing deployment cost and delay,and an enumeration algorithm and an approximate algorithm based on the improved entropy weight method and TOPSIS method are proposed to solve the problem of hierarchical deployment and load distribution h of edge datacenter.Finally,the simulation results show that the proposed two algorithms can minimize the deployment cost and delay cost,and the proposed hierarchical deployment scheme can effectively balance the deployment cost and delay.(2)A time feedback based task allocation and lightpath provisioning mechanism for coflow is proposed.In order to solve the problem of multi-coflow sharing computing and resource competition in the network resource scenario under the trend of "intensive edge interconnection",a mechanism of multi-coflow task allocation and lighpath configuration is proposed for multi-tenants sharing computing and network resources.First of all,it analyzes the correlation characteristics among the subtasks in the coflow task,considers the constraints of computing and network resources,establishes the mathematical model of the minimum completion time of a single coflow,and proposes a heuristic algorithm.Then,with the coflow completion time as an indicator,multiple coflow orders are arranged.Finally,according to the characteristic that the coflow completion time depends on the completion time of the slowest subtask,a mathematical model based on the coflow completion time is established to minimize the bandwidth consumption.At the same time,a heuristic algorithm of routing,modulation format and frequency slot configuration is proposed.The simulation results show that the coflow task scheduling and the lightpath configuration have important effects on reducing the delay and bandwidth consumption.(3)A multi-stage aggregation and resource allocation mechanism for dense coflow tasks is proposed.To solve the problem of distributed data processing under the trend of "dense optical interconnection",a joint optimization algorithm of multi-stage aggregation of distributed data and resource configuration is proposed.Firstly,by analyzing the influencing factors of the coflow completion time,the processing time model and communication time model are defined,and the completion time of cluster,completion time of stage and coflow completion time are formulated.Then,in order to optimize the number of stages of coflow aggregation,the number of clusters in each stage,the location of cluster center and the allocation of bandwidth resources,a heuristic algorithm of cluster center selection,cluster division and bandwidth allocation is proposed.Finally,the simulation results show that compared with the single-stage aggregation scheme,the proposed multi-stage aggregation scheme has the best performance in terms of delay and bandwidth in the scenarios of redundant data,good aggregation effect and large amount of geo-distributed data.