数据中心网络中自适应流量分配方案研究
2018-02-01吴杰徐昌彪
吴杰+徐昌彪



摘要:数据中心网络中的长数据流对网络流量的贡献很大,但是数量较少,因而容易产生网络拥塞,影响网络性能。为改善这一状况,提出一种基于软件定义网络(Software Definition Network,SDN)的多路径自适应流量分配方案。该方案利用SDN控制器的全局特性,计算长数据流的多条可用路径及其权值,并根据链路状态更新路径权值,实现长数据流的动态分配。仿真结果表明,与单路径方案相比,长数据流的吞吐量提升了28%,平均时延降低了16%;与等价多路径方案相比,长数据流的吞吐量提升了7%,平均时延降低了5%。相比于已有方案,该方案在提高网络吞吐量与降低网络时延方面都表现更好。
关键词:数据中心网络;自适应流量分配;路径全值;动态分配
DOIDOI:10.11907/rjdk.173235
中图分类号:TP391
文献标识码:A文章编号文章编号:1672-7800(2018)001-0179-05
Abstract:The long data flow in the network of data center contributes a lot to network capacity, but often leads to network congestion and damages network performance since it is less in quantity. In order to improve such condition, a multi-path adaptive flow allocation scheme based on software defined network is proposed. Depending on the global features of SDN controllers, it calculates multiple potential paths of a long data flow and their weights, and updates path weights with respect to link states for the purpose of dynamic allocation of long data flow. Simulations show that compared with the single path scheme, the throughput of long data stream is increased by 28%, and the average delay is reduced by 16%; compared with the equivalent multipath scheme, the throughput of long data stream is increased by 7% and the average delay is reduced by 5%. The program performs better in terms of network throughput enhancement and network delay decrease than existing programs.
Key Words:the network of data center; adaptive flow allocation; path weights; dynamic allocation
0引言
数据中心网络是数据传输、计算和存储的中心,集中了各种软硬件资源和关键核心业务,如Web服务、搜索引擎、MapReduce大规模集群计算等[1-4]。随着这些业务的迅速发展,现有TCP/IP数据中心网络的性能瓶颈日益凸显。软件定义网络(Software Definition Network,SDN)是一种全新的网络范式,其将数据平面与控制平面相耦合,实现逻辑集中控制。SDN具有网络管理简单灵活、可编程、虚拟化等优点,因此越来越多地被应用于数据中心网络[5-8]。
数据中心网络存在显著的长短流特征。文献[9]对6 000台服务器的数据中心网络一个月150TB的数据量进行统计分析后发现,不到20%的长流贡献超过80%的网络流量。无独有偶,文献[10]也观察到类似结论,指出约90%的网络流量由长流产生。……
