Research On Distributed Traffic Load Balancing Technology Based On SDN

Against the backdrop of the rapid development of the Internet,existing network architectures have gradually become unable to meet the application scenarios and usage requirements of rapidly increasing network traffic.Improving network performance has become a challenge that needs to be addressed by various service providers and data centers.Based on software-defined network architecture,this article designs and implements a distributed load balancing strategy based on network probes and congestion awareness in the programmable data plane.In addition,it further improves network transmission performance and resource utilization by combining with an in-band network telemetry framework and flow classification method.The main contributions of this article are as follows:(1)A load balancing strategy based on network probes was designed in the programmable data plane.This strategy addresses the problem of a single congestion-aware measurement metric by introducing multiple metric factors for load evaluation,which comprehensively measures the congestion status of the network.The optimal transmission path is selected by the network probe carrying the load evaluation factor to achieve end-to-end multipath traffic load balancing.Finally,based on the transmission characteristics of TCP traffic,the data flow is segmented into packet clusters,and business traffic is forwarded in a more granular way.(2)To address the problem of reduced available bandwidth caused by active sensing of link congestion using network probes,a method for automatically adjusting probe frequency based on end-to-end link state was designed by combining with an in-band network telemetry framework.This method obtains real-time information about the link through in-band network telemetry,calculates the load balancing difference of each link in the control plane,and adjusts the frequency of sending probes by edge switches through the load balancing difference.(3)To address the problem of the optimal path selected by probes quickly becoming congested,a flow classification method was designed by analyzing the basic distribution characteristics of traffic.This method dynamically adjusts the classification threshold by calculating the average available queue using queue information carried by network probes.The data flow is divided into large and small flows,and different forwarding methods are adopted for them.This effectively alleviates the problem of rapid congestion of the optimal path,and further improves the utilization of network resources for multiple reachable paths.Through simulation experiments,the load balancing strategy based on multiple load evaluation factors can achieve distributed traffic load balancing.Compared to HULA,it improves the average flow completion time performance by 1%～2% under high load(≧70%).The probe frequency adjustment method based on in-band network telemetry effectively reduces the additional bandwidth consumption of network probes.Compared to HULA and ULFC,it reduces it by 90% and 60% under low load(≦30%).Based on the probe-based load balancing,using flow classification method further improves network performance,reduces the congestion of the optimal forwarding path,and improves the average flow completion time performance by up to 67%,58%,35%,and 9% compared to ECMP,CONGA,HULA,and ULFC.