虚拟网络实验室支撑环境研究

【作者】 莫铁强；

【导师】 李仁发；

【作者基本信息】 湘潭大学 ， 电力电子与电力传动， 2002， 硕士

【摘要】 虚拟实验室的成功建立必须依托在能传输可靠的实时数据流的网络环境中。尽管人们对主机端的拥塞控制研究已取得了长足的进展，但目前的路由器队列管理机制不能提供端到端的拥塞控制机制以支撑这种网络环境，具体表现为发生拥塞时对不同类型流量的带宽分配或数据包丢失不具有公平性及网络中各TCP连接流量的同步性。因此，我们很有必要对传统的队列管理机制Drop_Tail、RED等进行改进以达到提高网络服务质量的目的。本文分析了主机端的拥塞控制机制和网络中路由器的现有各种队列管理机制：绝对优先级队列机制、加权公平队列(Weighted Fair Queue)[1]机制、基于类的队列(Class-based Queue)[2]机制以及RED(Random Early Detection)[3]机制的主要特点，并重点针对IETF(Internet Engineering Task Force)推荐的把RED算法与明确的拥塞通告ECN(Explicit Congestion Notification)[4]结合的方法进行了分析，发现随着Internet流量的日益递增，RED机制的依赖平均队列长度来管理拥塞控制的算法并不能有效地阻止包丢失：与Drop-Tail相比，RED确实消除了对阵发流量(TCP)的偏见，但是增加了平滑流量(UDP)的丢包概率。因此本文在分析各种RED机制的基础之上，对RED算法作了一定的改进，提出了一种根据带宽占有量而把各种流量分类且称之为New_RED的机制,这样就保证了网络中各连接流量共享瓶颈带宽。最后本文就RED机制和New_RED机制的性能利用Berkeley的NS(Network Simulator)仿真器[5]进行了仿真性能的对比，实验表明New_RED的丢包性能得到了一定的改进。更多还原

【Abstract】 The successful setup of Virtual Laboratory will be dependent on reliable real-time data stream transmission in the network environment. Although the congestion control research of host-ends has made great strides,the current queue management mechanism in router is not able to provide end-to-end congestion control mechanism to sustain the above environment, namely, the unfairness of assigned bandwidth or drop-rate for different types of flows and the synchronization of all TCP flows in network. So it is necessary to improve the classical queue management mecnanism such as Drop_Tail and RED in order to boost the QoS(Quality of Service ). This article analyzes the main traits of host-ends congestion control and all kinds of fashionable queue management machanism ,which include Queue Mechanism with absolute priority ,Weighted Fair Queue Mechanism and Class-based Queue Mechanism . Further , we emphatically analyze the method of combined RED and ECN which is recommended by IETF . In the analysis we find that with the increase of Internet traffic ,RED mechanism doesn’t effectively prevent the packets from dropping . In comparison with Drop_Tail Algorithm , RED Algorithm surely eliminates the bias against the bursty traffic(TCP),but increases the drop probability of smooth traffic(UDP) . So on the bases of the analytic RED mechanism , we make an improvement in maintaining the fair bandwidth shares among all flows in bottleneck links,which we call New_RED Algorithm and classfies all flows according to bandwidth shared by per flow. Finally the article also compares the simulated performance between RED mechanism and New_RED mechanism by the use of NS . The simulated results verify that New_RED mechanism will work better.更多还原