华南理工大学学报(自然科学版) ›› 2006, Vol. 34 ›› Issue (6): 89-94.

• 计算机科学与技术 • 上一篇    下一篇

基于网络测量和模糊控制技术的拥塞控制机制

宋丽华1 王海涛2 陈鸣1   

  1. 1.解放军理工大学 指挥自动化学院,江苏 南京 210007;2.解放军理工大学 通信工程学院,江苏 南京 210007
  • 收稿日期:2005-01-21 出版日期:2006-06-25 发布日期:2006-06-25
  • 通信作者: 宋丽华(1976-),女,博士生,主要从事网络性能、网络测量方面的研究 E-mail:minnehaha@126.com
  • 作者简介:宋丽华(1976-),女,博士生,主要从事网络性能、网络测量方面的研究
  • 基金资助:

    国家自然科学基金资助项目(90304016)

Congestion Control Scheme Based on Network Measurement and Fuzzy Logic Control Technology

Song Li-hua1  Wang Hai-tao2 Chen Ming1   

  1. 1.Institute of Command Automation,PLA Univ.of Sci.and Tech.,Nanjing 210007,Jiangsu,China;2.Institute of Communication Engineering,PLA Univ.of Sci.and Tech.,Nanjing 210007,Jiangsu,China
  • Received:2005-01-21 Online:2006-06-25 Published:2006-06-25
  • Contact: 宋丽华(1976-),女,博士生,主要从事网络性能、网络测量方面的研究 E-mail:minnehaha@126.com
  • About author:宋丽华(1976-),女,博士生,主要从事网络性能、网络测量方面的研究
  • Supported by:

    国家自然科学基金资助项目(90304016)

摘要: 为向端系统反馈及时准确的网络状态以提高TCP效率,同时避免主动队列管理(AQM)机制的诸多局限,提出一种基于网络测量的拥塞控制机制——利用分布在网络中的测量设施监测主干链路,再根据链路性能采用模糊控制技术指导端系统选择适当的FAST控制参数.仿真实验表明,该机制能承受更大的负载压力,达到高吞吐量,并能稳定排队时延,在高速网络中比AQM机制更稳定、更公平.相比突发性的Web流量来说,该机制更适用于类似P2P等数据量大、连接持续时间长的流量的拥塞控制.

关键词: 网络测量, 模糊控制, 拥塞控制, 吞吐量, 排队时延

Abstract:

A network-measurement-based congestion control scheme is proposed to avoid the limitations resuhing from the AQM (Active Queue Management)mechanism and improve the TCP performance by feeding back timely and correct information on network states to end systems.In the proposed scheme,the measurement facilities dis.tributed in networks ale exploited to acquire the perform ance of network backbone,an d according to the resulting data a fuzzy logic controller is used to help end systems determ ine the appropriate FAST control parameters. Simula-tion experiments indicate that the proposed scheme can support extremely heavy load,achieve high throughput,and steady the queuing delay.Moreover,it behaves more stably and fairly than the AQM scheme in high-speed networks,and is more suitable for P2P like large-volume long.lived flows rather than Web.like bursty traffic.

Key words: network measurement, fuzzy logic control, congestion control, throughput, queuing delay