Research And Implementation Of Elephant Flow Load Balancing In Data Center Network Based On SDN

With the rapid development of the internet in recent years,the traditional network model can no longer meet the needs of users.Therefore,it is necessary to explore and discover new network transmission modes for users to use.Software Defined Network(SDN)is very good.To solve this problem;the biggest difference between SDN network and traditional network lies in the use of control and forwarding separation technology to achieve network transmission,through the use of Openflow protocol to achieve network data traffic scheduling;in SDN network,the control layer is the most important one in this layer,not only can provide an application program interface for the application layer,but also can flexibly schedule the data information in the data layer;due to the huge data traffic in the data center network,it is necessary to introduce new technologies to improve the traffic in the network distribution efficiency and forwarding efficiency,so this paper introduces the SDN network into the data center network,and applies it to the data center network through the idea of separation of control and forwarding.Based on the above problems of data transmission in the data center network combined with the advantages of the SDN network,this paper proposes the traffic identification and overall load balancing framework to solve the current large data flow network load imbalance problem.Firstly,introducing the current development status of the SDN network and the current status of the data center network,and introduce the SDN network architecture,listing the advantages and disadvantages of the load balancing algorithm proposed by the SDN-based data center network in recent years,and summarize the shortcomings by comparison and find out in the data center network,there is an inaccurate phenomenon in the identification of data traffic,that is,the simultaneous transmission of large and small streams may cause uneven transmission of the network,so a two-stage identification method for elephant flows is proposed: in the first stage,the missed detection rate and false detection consider the rate comprehensively,and set a threshold to use sFlow to sample to identify suspicious elephant flow;in the second stage,identify the real elephant flow by setting the tolerance value of the flow duration.Secondly,applying the identified elephant flow to the overall load balancing strategy,and dividing the overall load balancing framework into a network-wide topology awareness module,link state collection module,elephant flow identification module,routing decision module,congestion processing and flow The table forwarding module uses the link layer discovery protocol to obtain the network topology in the network topology awareness module,and uses the k shortest path algorithm in the routing decision module to calculate the k shortest paths,and then uses the fuzzy comprehensive evaluation algorithm to evaluate the best among the k paths a path is a preferential forwarding path.The congestion processing and forwarding module adjusts the forwarding order of the paths according to the network information collected by the link state collection module.Multiple modules coordinate with each other to complete the load balancing of the overall network.Finally,it is verified by experimental simulation,and the network throughput,transmission delay,and link utilization are used as performance evaluation indicators to verify the elephant flow detection mechanism and the overall load balancing strategy proposed in this paper.The verification results show that the proposed elephant flow load balancing the strategy is better than the commonly used ECMP and Hedera algorithms,which effectively improves the efficiency of elephant flow detection and provides a better elephant flow load balancing strategy for the data center network.