基于流分类的数据中心网络负载均衡机制  

Load Balancing Based on Flow Classification for Datacenter Network

作　　者：崔子熙 胡宇翔[1] 兰巨龙[1] 王雨 CUI Zi-xi;HU Yu-xiang;LAN Ju-long;WANG Yu(Information Engineering University,Zhengzhou,Henan 450002,China;Guangdong Communications & Networks Institute,Guangzhou,Guangdong 510670,China)

机构地区：[1]信息工程大学,河南郑州450002 [2]广东省新一代通信与网络创新研究院,广东广州510670

出　　处：《电子学报》2021年第3期559-565,共7页Acta Electronica Sinica

基　　金：国家自然科学基金资助项目(No.61521003,No.61872382);国家重点研发计划课题(No.2017YFB0803204);广东省重点领域研发计划项目(No.2018B010113001)。

摘　　要：为充分利用数据中心网络的多路径带宽,现有研究多采用基于链路感知的负载均衡算法,在动态获取全局链路拥塞信息后选取最优路径对流量进行转发.然而这些研究未考虑数据中心网络流量大小分布不均匀的特性,难以在选路成本和转发效率上取得平衡.为此,设计一种基于流分类的数据中心网络负载均衡机制(ULFC,Utilization-aware Load balancing based on Flow Classification),在实现拥塞感知的基础上进行流量特征分析,采用不同的策略为大、小流分配路径,实现网络流量特征与选路方法优势的最佳匹配.实验结果表明,相比于现有方案,ULFC的平均流处理效率提高了1.3倍至1.6倍,路由成本降低了50%以上.In order to fully utilize the bandwidth of multi-paths of the datacenter network(DCN),the existing studies mostly adopt the congestion-aware load-balancing scheme,which forwards traffic along the optimal path after dynamically obtaining global congestion information.However,these works do not consider the non-uniform distribution of flow size and are difficult to strike a balance between the routing cost and the forwarding efficiency.This paper proposes ULFC,a utilization-aware load-balancing mechanism based on flow classification.By analyzing the characteristics of traffic,ULFC classifies the flows based on their sizes and assigns paths to them using different strategies,realizing the best matching between the characteristics of traffic and the advantages of the routing method.We evaluate ULFC with simulation and the results show that it outperforms the existing schemes in average flow-completion time(1.3~1.6×),while the routing cost has been reduced by more than 50%.

关 键 词：数据中心网络 负载均衡 可编程数据平面 流分类 可编程协议无关报文处理 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]