Research On Data Transfer Scheduling In Inter-Datacenter Network

With the rapid development and large-scale application of online services such as geo-distributed big data analysis and content distribution,the inter-datacenter WAN traf-fic is growing explosively.Compared with the intra-datacenter network,WAN bandwidth resources are very scarce.Unreasonable bandwidth resource allocation will cause appli-cation performance loss and bring high bandwidth cost.The existing transfer schedul-ing schemes for inter-datacenter traffic mainly focus on the Point-to-Point(P2P)trans-fer requirements,which can not be well applied to the data transfer requirements with Multi Point-to-Point(MP2P)and Point-to-Multi Point(P2MP)communication modes,and thus are difficult to obtain the best transfer performance.In view of the shortcomings of the existing work,this thesis focuses on the MP2 P and P2 MP transfer scheduling problem in inter-datacenter network,proposes efficient transfer scheduling algorithms considering the traffic characteristics and network characteristics,and verifies the effectiveness of the algorithms through experiments.The main contents of this thesis are as follows:(1)For the MP2 P aggregation transfer requirements,this thesis studies the charac-teristics of this kind of aggregation flow and proposes a hierarchical aggregation scheme.Firstly,the transfer scheduling problem of single transfer request is modeled as a mixed integer linear programming problem,aiming at minimize the completion time of MP2 P transfer,and the complexity of the problem is analyzed.On this basis,a transfer schedul-ing algorithm 2AGG based on linear programming relaxation technique and random round-ing technique is proposed to solve the problem efficiently.Furthermore,an online multi-request joint scheduling algorithm 2AGG-M is proposed.Experimental results show that2 AGG and 2AGG-M can reduce the transfer completion time of MP2 P traffic by 30% -70% compared with the existing algorithms.(2)For the P2 MP distribution transfer requirements,this thesis studies the character-istics of this kind of distribution flow and proposes a traffic distribution scheme based on multicast routing.Firstly,the mathematical model of the problem is established,aiming at minimize the cost of network bandwidth,and the complexity of solving the problem is analyzed.On this basis,this thesis proposes an online P2 MP traffic scheduling algorithm CDScheduler.By using the step-by-step problem solving scheme,CDScheduler can min-imize the cost of bandwidth and improve the network throughput to speed up the comple-tion of every transfer request.This thesis compares the CDScheduler algorithm with the existing algorithms.The experimental results show that compared with the comparison algorithm,the CDScheduler algorithm can effectively reduce the network bandwidth cost by up to 70%,and improve the bandwidth utilization.