首页|面向多模态网络的SDN交换机流量调度方法

面向多模态网络的SDN交换机流量调度方法

扫码查看
多模态网络需要识别和解析多种新型模态流量,以实现面向流表的定制化调度.SDN交换机的数据包调度能力决定整个系统的负载分配,对于系统性能起着至关重要的作用.通常情况下采用网卡提供的分流策略将数据包分发至不同的CPU核,但网卡分流策略仅适用于支持的数据包类型,难以满足多种新型模态流量的解析需求.针对上述问题,本文提出一种用于多核SDN交换机的支持预分类和适配分流策略加载的流量分发调度方法.通过协议无关的预分类确定模态所属的流水线集合,同时支持为不同模态加载适配分流策略,将模态流量均衡调度至对应流水线.在基于DPDK的POF交换机中实现了本文提出的调度方法,实验结果表明,相比原生POF交换机,新方法在处理多模态流量时提升约1.54倍的系统吞吐性能,并有效地降低了时延.
SDN Switch Traffic Scheduling Method for Polymorphic Network
Polymorphic network needs to recognize and parse various new modality traffic to achieve stream-oriented customized schedu-ling.The packet scheduling capability of SDN switches determines the load distribution of the entire system,which plays a crucial role in system performance.In general,the packet distribution to different CPU cores is carried out by the diversion strategy provided by the network card,but the problem is that the diversion strategy of NIC is only suitable for supported packet types and cannot meet the par-sing requirements of various new modality traffic.To address the above issues,this paper proposes a traffic distribution scheduling method for multi-core SDN switches that supports pre-classification and the loading of adaptive diversion strategies.By using protocol-agnostic pre-classification to determine the pipeline set to which the modality belongs,it also supports loading adaptive diversion strate-gies for different modalities,evenly scheduling modality traffic to the corresponding pipeline.The experiment implemented the proposed scheduling method in a DPDK-based POF switch,and the results show that compared with the original POF switch,the new method has improved the system throughput performance by about 1.54 times when dealing with multimodal traffic and has effectively reduced la-tency.

polymorphic networktraffic schedulingload balanceSDNswitch

徐洁、刘磊、吴京洪

展开 >

中国科学院声学研究所 国家网络新媒体工程技术研究中心 北京 100190

中国科学院大学 北京 100049

鹏城实验室 深圳 518055

多模态网络 流量调度 负载均衡 软件定义网络 交换机

国家重点研发计划宽带通信与新型网络应用示范课题

2020YFB1806402

2024

网络新媒体技术
中国科学院声学研究所

网络新媒体技术

CSTPCD
影响因子:0.208
ISSN:2095-347X
年,卷(期):2024.13(3)
  • 7