基于SDN网络大象流负载均衡算法研究

Research on Elephant Flow Load balancing Algorithm Based on SDN

摘要:

摘 要：负载均衡算法是通过对网络中的流量进行调度来提高网络资源利用率，是计算机网络中的一个重要研究方向；针对网络中大象流导致的网络拥塞和老鼠流的排队时延等负载不均衡问题，提出了带宽和时延加权负载均衡（BD-WLB）算法来提高负载均衡性能，综合考虑了大小流之间的流量特征不同，改进了传统算法的路径计算方式；算法通过控制器来获取网络流量和状态信息；然后利用带宽和时延等网络状态参数来为大象流和老鼠流分别计算最优路径；采用P4语言来对数据平面转发流程进行优化处理；实验结果表明，在高负载状态时，BD-WLB算法相比于ECMP算法提高了38.4%的网络吞吐量和41.9%的链路利用率，降低了41.8%的网络时延；使网络资源得到了更好的利用，证明了BD-WLB算法的可行性和有效性；

Abstract:

Load balancing algorithm is an important research direction in computer network, which can improve the utilization of network resources by scheduling the traffic in the network. In order to solve the load imbalance problems such as network congestion caused by elephant flow and queue delay of rat flow in the network, the bandwidth and delay weighted load balancing (BD-WLB) algorithm is proposed to improve the load balancing performance. The path calculation method of the traditional algorithm is improved by considering the different traffic characteristics between small and large streams. The algorithm obtains network traffic and state information through the controller. Then the optimal path is calculated for elephant flow and mouse flow respectively by using network state parameters such as bandwidth and delay. P4 language is used to optimize the data plane forwarding process. The experimental results show that the BD-WLB algorithm improves the network throughput by 38.4%, the link utilization by 41.9% and the network delay by 41.8% compared with the ECMP algorithm at high load. It makes better use of network resources and proves the feasibility and effectiveness of BD-WLB algorithm.