数据中心网络流量分类路由机制研究

Classified Flow Routing Scheme for Data Center Networks

当前数据中心网络数据流量大小分布不均衡,传统等价多路径转发ECMP算法容易将多条大数据流转发至同一链路,导致链路瓶颈.提出面向CLOS结构数据中心网络的基于SDN的流量分类路由SCR(SDN-Based Classified Routing)机制.SCR利用Open Flow机制周期性统计数据流来计算阈值,通过动态判定方法将数据流分为大流和小流.大流由SDN控制器通过自适应路由算法计算路径,小流由交换机通过流量无视路由算法计算路径.本文在VL2架构上建立合成流量模型对SCR进行性能分析与评价.仿真与分析结果表明,与ECMP相比,SCR在网络吞吐量、数据流丢弃率、分组端到端时延和平均队列长度等方面具有优势.

In the current data center networks, the flow size distribution is not uniform. The key problem of Equal-Cost-Multi-Path （ ECMP ） mechanism is that large,long-lived flows traversing a switch may be mapped to the same output port resulting in collisions. Such ＂collisions＂ can cause load imbalances across multiple paths and network bottlenecks. To address this issue, we present SDN- Based Classified Routing （ SCR ） scheme for Clos Data Center Networks. SCR adopts the centralized OpenFlow controller approach for network management. SCR classifies flows into elephant flows and mice flows. The classification threshold is calculated on the controller periodically. The elephant flow is scheduled by the adaptive routing algorithm on the controller, and the mice flow is sched- uled by the oblivious muting algorithm on switches. We evaluate the proposed SCR scheme through simulations on VL2 network com- pared with ECMP. The results show that SCR has better performance in the network throughput, flow discard rate, packet delay and the average queue length compared with ECMP.