摘要
通过对比分析reno、vegas和fast等几种重要的传输控制协议(TCP)拥塞控制机制,验证了fast的高性能,同时针对其缺少全局状态、控制参数不易设置的缺陷,提出了一种基于测量的改进方案.基本思想是:利用测量技术周期性监测网络主干链路,根据链路性能为竞争连接设定适当的期望冗余包参数,在不依赖路由器的条件下使瓶颈排队趋向于一个固定长度.仿真实验表明,改进后的fast能够获得更高、更稳定的power值,其power值不随竞争连接增多而下降.
Congestion control is one of the cornerstones of Internet. The classical transmission control protocol (TCP) congestion control scheme, Reno, and the other two, Vegas and Fast, recently suggested upgrades for high-speed networks, are analyzed. In the comparative work, Fast's high performance is validated and its weakness, the difficulty in setting parameters, is revealed. Also, to facilitate Fast parameter setup, a measurement-based improvement is proposed. The essential idea is to introduce a macroscopic guidance layer upon end systems which will guide them to appropriate settings of number of extra packets according to periodical performance measurement on network backbone. Simulation results indicate that by realizing the target of having the bottleneck queue tend to be a fixed length, the improved Fast scheme can get a higher and steadier power, and its power will not decrease when new connections join in.
出处
《北京邮电大学学报》
EI
CAS
CSCD
北大核心
2005年第4期27-31,共5页
Journal of Beijing University of Posts and Telecommunications
基金
国家"863计划"项目(2001AA112090)
国家自然科学基金重大研究计划项目(90304016)
关键词
网络测量
拥塞控制
传输控制协议fast
冗余包
network measurement
congestion control
transmission control protocol fast
extra packets