【摘要】 针对软件定义广域网(software-defined wide-area networks,SD-WAN)对数据传输时延要求高而有必要量化评估其分组转发性能的问题，考虑不同类型分组处理过程的差异性，基于排队论构建1个精确的OpenFlow分组转发性能模型。首先，描述SD-WAN网络的典型部署场景，并对其数据平面的到达分组进行分类，进而为OpenFlow交换机的分组处理过程构建M/H3/1排队模型，并推导其关键性能指标的表达式。在此基础上，结合控制器集群的Packet-in消息排队模型，建立面向SD-WAN网络的OpenFlow分组转发性能模型，进而求得平均分组转发时延的闭式解。最后，采用模拟仿真和数值分析方法对本文所提出的模型进行评估。研究结果表明：与现有模型相比，本文所提出的模型能更准确地估计不同类型分组在OpenFlow交换机中的平均逗留时间，且平均分组转发时延主要受旧流分组处理速率和新流分组概率的影响。更多还原

【Abstract】 It is vital to quantitatively evaluate the packet forwarding performance of software-defined wide-area networks(SD-WAN) due to their strict requirements on data transmission delay. An accurate OpenFlow-based packet forwarding performance model was constructed based on queueing theory considering the processing diversity of different types of packets. Firstly, typical SD-WAN deployments were depicted, and arrived packets in their data planes were classified. Then, an M/H3/1 queueing model for the packet processing of OpenFlow switches was built and the expressions of its key performance metrics was deduced. Furtherly, a performance model of OpenFlow-based packet forwarding in SD-WAN were established in combination with that of packet-in messages in controller clusters, and the closed-form solution of its average packet forwarding delay was derived.Finally, the simulation and numerical analysis methods were applied to evaluate the proposed model in the paper by experiments. The results show that the proposed model in the paper can more accurately estimate the average sojourn time of different types of packets compared with the existing models, and the average packet forwarding delay is chiefly affected by the processing rate of packets in old flows and the probability of a packet belonging to a new flow.更多还原