The Pseudo Congestion of TCP in Software-Defined Networking
-
摘要: 软件定义网络(software-defined networking, SDN)将控制平面与转发平面分离，通过控制器配置交换机流表项来实现网络的灵活控制，极大地提高了网络带宽利用率.随着SDN的蓬勃发展，越来越多的高校和公司开始部署SDN.同时SDN也面临着一些传统IP网络中不存在的问题，如一些原本在IP网络中运行良好的协议在SDN网络中性能却受到了严重的影响，TCP协议就是其中之一.从SDN的工作机制出发，通过3个场景阐明了SDN在proactive工作模式下依然存在Packet-In短时间内高速并发的可能性.总结并实验验证了高速并发的Packet-In以及流表更新时旧表项需重排列的特性都会使数据包在SDN网络中产生较大时延.实验结果表明，当TCAM支持4000个流表项时，最坏情况下仅由新插入流表项优先级原因导致已有流表项的重排列就能使单次传输时延达到10s，并发的高速Packet-In则会使时延加大.以实验为基础，揭示了由于SDN网络特性造成的伪拥塞现象，即传统TCP在SDN网络中面临两大问题：1)TCP建立连接困难；2)TCP协议传输低效.最后通过对实验现象进行分析，提出了解决SDN网络中TCP低效问题可能的工作方向.Abstract: software-defined networking (SDN) separates the control plane and the data plane, and this kind of separation can achieve flexible control via deploying fine-grained rules on the flow tables in switches, while potentially improving the utilization of network bandwidth. With the development of SDN, more and more campus and enterprise network begin to deploy network based on SDN. During this procedure, SDN has encountered some problems which dont exist in the traditional IP network. For example, some protocols used in the existing IP network are subject to great challenge in SDN based network, such as TCP, which is the most basic protocol in TCP/IP network. First, we make a penetrating analysis on the working mechanism of SDN, and three examples are given to illustrate that it is quite possible to generate large volume of Packet-In messages even in proactive mode. Then experiments are carried out and the results show that the end-to-end TCP connections have experienced a large delay caused by the SDN unique operations such as re-organizing of rules in TCAM and fast Packet-In message generating. In the worst case, the delay caused by the reordering of the rules can reach up to 10 seconds when the TCAM contains 4000 flow entries in our experiments. Based on the experimental results, we highlight two major problems when applying traditional TCP protocol in SDN networks: one is that it is hard to establish the connection, and the other is the transmission inefficiency. Through the analysis of the experimental results, we propose the possible direction to solve TCP inefficiency issue in SDN.
-
-
期刊类型引用(7)
1. 黄玲，黄镇伟，黄梓源，关灿荣，高月芳，王昌栋. 图卷积宽度跨域推荐系统. 计算机研究与发展. 2024(07): 1713-1729 . 
本站查看
2. 杨玲玲. 基于HM与LWR算法的电子设备MCS推荐优化. 山西电子技术. 2024(04): 22-24 . 百度学术
3. 郑升旻，付晓东. 利用混合Plackett-Luce模型的不完整序数偏好预测. 计算机应用. 2024(10): 3105-3113 . 百度学术
4. 杜兆芳. 基于协同排序学习算法的移动群智感知任务推荐. 电子产品世界. 2023(09): 64-66+70 . 百度学术
5. 朱丽丽. 随机森林算法下列表级排序学习推荐系统设计. 淮阴工学院学报. 2023(05): 62-68 . 百度学术
6. 曹玉红，赵乙，陈佳桦. 兼容异构数据的稳定评估模型. 小型微型计算机系统. 2021(09): 2011-2016 . 百度学术
7. 林子楠，刘杜钢，潘微科，明仲. 面向推荐系统中有偏和无偏一元反馈建模的三任务变分自编码器. 信息安全学报. 2021(05): 110-127 . 百度学术
其他类型引用(4)
计量
- 文章访问数: 1217
- HTML全文浏览量: 2
- PDF下载量: 759
- 被引次数: 11
下载:
