Research On SDN Data Center Flow Scheduling Algorithm Based On Link State

Recent years,the continuous development of network technology has promoted the vigorous development of several fields,including big data,cloud computing and network communication services,the expanding of the scale of networks in data centers,the growth of the number of network services needed to be processed and network flow,which puts forward higher requirements for the flow scheduling of data centers.However,traditional flow scheduling algorithms often cannot think from the overall situation,and do not fully consider the impact of the real-time state of the link on flow transmission,and also cannot schedule the explosive growth of network traffic in nowadays well.In recent years,the proposal of Software Defined Network(SDN)has provided new ideas for the study of flow scheduling problems.Based on the analysis of the SDN architecture,with the help of SDN’s centralized control and network programmable advantages,this thesis is able to collect the real-time status of network links.And considering elephant flow in the data center traffic requiring of high bandwidth,and mouse flow being sensitive to delay,this thesis combines Link status and flow characteristics,uses ant colony algorithm for flow scheduling,and proposes a SDN data center traffic scheduling algorithm(DCFS-LS)based on link status.First,this thesis analyzes and studies the link state parameters in the SDN-based data center network,interacts with Open Flow messages in control of plane cyclicity through bottom network of SDN to complete the collection and calculation of network topology and real-time link status information,and divides the data flow based on the ratio of the rate to the link bandwidth into two types: elephant flow and mouse flow.Secondly,this thesis analyzes and designs the SDN data center flow scheduling algorithm DCFS-LS based on link status.By analyzing the characteristics of link state parameters and dataflow,several important influencing factors of the flow scheduling algorithm are determined,including available bandwidth,rate,delay,and elephant and mouse flow categories,then the ant colony algorithm finding the global optimal solution can be modified and optimized,and can be used to schedule elephant flow and mouse flow.The available bandwidth is set as the initial pheromone on the link to avoid the problem of too strong randomness at beginning elephant and mouse flow scheduling of link selection by ant.The ratio of the available bandwidth to the link bandwidth is set as the heuristic function of the elephant flow,and the reciprocal of the delay as the heuristic function of the mouse flow,then the thesis uses the link pheromone,heuristic function of the elephant flow and the mouse flow to calculate the transferring probability of the flow selecting the next node,which guides the ant to to choose the path with a lighter load for a elephant flow and a path with a lower delay for a mouse flow.Then,the functional modules of the DCFS-LS algorithm and the underlying network topology are designed and implemented.The functional modules mainly include topology awareness module,network monitoring module,scheduling module and flow table management module.In the Ryu controller,Python programming is used to implement related functions and algorithm deployment.Finally,a simulation experiment is carried out on Mininet.The DCFS-LS algorithm in this thesis is compared with the ECMP algorithm and the Hedera algorithmin the aspects of average link utilization,average throughput and average transmission delay.Experiments show that the DCFS-LS algorithm reduces the average transmission delay,improves average link utilization and average throughput when there is a lot of traffic or a high ratio of elephant flow between Pods.