基于SDN-DDQN的数据中心网络负载均衡算法

Data Center Network Load Balancing Algorithm Based on SDN-DDQN

在数据中心网络(DCN)数据流量激增、大小流突发的情况下,采用传统负载均衡算法存在实时性不足,长期效果难以优化等问题,易造成网络链路拥塞。为此,提出一种基于SDN-DDQN的负载均衡(DDQNLB)算法。此算法利用SDN全局视图的优势,选择交换机负载和带宽利用率作为网络状态输入,为DCN中的大流和小流分别设置卷积神经网络(CNN)进行学习和训练,以满足DCN中大流高吞吐量和小流低延时的需求。实验结果表明,与ECMP和Hedera算法相比,DDQNLB算法可有效提高网络吞吐量并保证较低的丢包率。

In response to the surge of data flow and the sudden occurrence of large and small flows in data center network(DCN),traditional load balancing solutions lack real-time responsiveness and long-term optimization,resulting in network congestion.A load balancing algorithm called DDQNLB based on SDN-DDQN is proposed.This solution takes advantage of SDN′s global view and selects switch load and bandwidth utilization as network state inputs.Convolutional neural networks(CNN)are set up for learning and training the large and small flows in DCN,respectively,to meet the high throughput of the large flows and low latency requirements of small flows in DCN.The experimental results show that compared with ECMP and Hedera algorithms,DDQNLB can effectively improve network throughput while maintaining a lower packet loss rate.