Network Load Balancing System Based On In-band Network Telemetry

In recent years,driven by the rapid development of new-generation information technologies such as cloud computing,big data,and artificial intelligence,data centers,as the infrastructure for massive data storage and computing,carry various online application systems.As the scale of network users continues to grow,the network traffic inside the data center is getting higher and higher,and the transmission performance of the data center network is facing huge challenges.This paper mainly studies the load balancing method of the data center network.Based on the software-defined network architecture and in-band network telemetry technology,the load balancing system of data center network is designed and implemented.The controller is used to realize traffic measurement and network monitoring,and realize dynamic multipath load balancing.First of all,this paper proposes a simple,general,and easy-to-deploy traffic measurement method.The traffic measurement function is realized on the P4 switch based on the Bloom Filter and the Count-Min Sketch structure,and the measurement data is reported regularly and actively through the INT-XD technology,and the control plane stores the measurement data into the key-value database to realize the measurement of all active data flows,reduce the memory overhead of the switch,and improve the accuracy of flow measurement.Then,according to the application requirements,traffic characteristics and topology structure in the data center network scenario,this paper proposes a load balancing system based on flow path isolation,and isolates the transmission paths of elephant flow and mouse flow based on the M/G/l queuing model.To meet the completion deadline of the mouse flow,design a corresponding traffic scheduling algorithm to improve network transmission performance.Finally,this paper implements a network load balancing system based on the Bmv2 software switch,and then verifies the relevant performance indicators through comparative experiments.The experimental results show that the traffic measurement method proposed in this paper has higher accuracy under the premise of the same switch memory overhead,and the load balancing system can effectively optimize the completion time of the mouse flow and the throughput of the elephant flow.