基于Gilbert丢包机制的TCP吞吐量模型  

A TCP Throughput Model Based on Gilbert Packet Loss Pattern

作　　者：曾彬[1] 张大方[2] 黎文伟[2] 谢高岗[3] 

机构地区：[1]湖南大学计算机与通信学院,湖南长沙410082 [2]湖南大学软件学院,湖南长沙410082 [3]中国科学院计算技术研究所,北京100190

出　　处：《电子学报》2009年第8期1728-1732,共5页Acta Electronica Sinica

基　　金：国家973重点基础研究发展计划(No.2007CB310702);国家自然科学基金重大研究计划(No.90718008);国家自然科学基金(No;60673155;No.60703097)

摘　　要：丢包机制是推导TCP吞吐量模型的关键,直接影响模型的准确性.本文利用四状态Gilbert丢包机制来描述端到端路径上的丢包行为,对TCP的拥塞控制过程进行建模,在此基础上提出了一种更精确的TCP吞吐量模型.实验表明,改进的模型能较好的与实际值相拟合,可以更精确地预测实际TCP数据流的吞吐量性能.TCP throughput is one of the key concepts of computer network, and its estimation is very important for many research and application domains. Packet loss pattern is crucial to deduce a TCP throughput model, and can affect model＇s precision directly. Therefore, a new TCP throughput model GT（Gilbert Throughput ）is presented by modeling a TCP congestion control mechanism based on Gilbert four-state model that is used to represent the packet lost behavior of end-to-end intemet path.Experiment resuits show that GT matches the results better than Goyal model, and can predict the throughput of TCP flows more precisely in actual network.

关 键 词：网络测量 TCP吞吐量 拥塞控制 Gilbert丢包机制 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]