【作者】 陶小梅；

【导师】 陈元琰；

【作者基本信息】 广西师范大学 ， 计算机软件与理论， 2005， 硕士

【摘要】 随着计算机网络的迅速发展,互联网用户数量激增,多媒体数据流迅速增长,网络的拥塞问题变得越来越严重。拥塞是一种复杂的现象,拥塞控制也是一个复杂的课题,而拥塞控制是确保Internet鲁棒性(robustness)的关键因素,因此网络的拥塞控制问题成为目前关于Internet研究的一个热点问题和难点问题。目前TCP是数据传送所使用的最为广泛的传输协议,它是一个端到端协议,将中间的网络视为一个黑箱,它没有直接确定网络情况的手段,而必须在与其它端系统交互的基础上推断出网络的情况。TCP协议的拥塞控制机制都是基于和式增加乘式减少(AIMD)算法, 因此该协议在高带宽时延乘积的环境下表现出来的性能非常糟糕。根据TCP存在的问题和高带宽时延乘积网络的特点,美国麻省理工大学的Dina Katabi提出了一种新的Internet拥塞控制框架,该协议称为XCP(eXplicit Control Protocol),显式控制协议。XCP是一种联合端系统和路由器共同协作的协议,将拥塞控制从带宽分配策略中解耦。实验证明XCP比TCP在高带宽时延乘积网络的环境中有着更好的性能,在效率性、公平性、灵活性以及稳定性方面表现都非常出色。甚至是在传统的环境中,XCP也能提高网络带宽的利用率,减小队列长度,使丢包的情况降至极低。通过将XCP协议的结构及执行过程的详细分析,并将其性能、稳定性及可行性与其它协议进行比较,我们认为XCP具有很多优点,是一种设计合理,考虑较全面细致并且可行的协议,是未来高带宽时延乘积网络协议中的强有力竞争者,因此我们选择XCP协议进一步研究改进。改进的工作主要有两个方面:1. 在接收端添加许可控制。由于在高带宽时延乘积网络中网络的传输速度高达每流Gbit/s,若接收端的计算机不能快速的处理高速到达的分组而产生了丢包,那么将会大大降低了网络性能,这时接收端主机成为传输了瓶颈。因此我们认为接收端的接收能力也应该列入考虑的范围之内,我们针对接收端进行了结构调整,在接收端添加了许可控制。接收端可以根据实际情况标识自身的接收能力并通过XCP原有的Feedback字段反馈给发送端,使发送端对发送速率做出相应的调整,避免不必要的丢包。我们通过模拟实验以及稳定性分析,证明在接收端添加许可控制并不影响系统的稳定性,是可行的。改进后使XCP的协议框架更完善,Feedback值中的信息包括了发送端请求增长的吞吐量、网络的拥塞情况和以及接收端的接收能力。而且还将流量控制的部分功能集成到了拥塞控制当中,使得流量控制有一种新的更简单的方式解决问题,同时减少窗口字段的使用。2. 通过对参数α的自适应调整提高网络的利用率。根据我们对XCP的参数α、β和γ进行的实更多还原

【Abstract】 With the rapidly improvement of computer network, the increase of Internet users, the popularity of multimedia flow, problem of Internet congestion control becomes more and more serious. Congestion control is the key factor to assure the Internet robustness. It becomes the hot issue and difficulty issue of the Internet research.. TCP as the most wildly used transmission protocol, it is a kind of end-to-end protocol. It takes the network or the Internet as a black box and can’t know the network status directly. It only can deduce the network status by end system . Current TCP_based congestion control exhibits poor performance in environments with a large per-flow bandwidth-delay product . As a remedy, Dina Katabi propose the eXplicit Control Protocol (XCP), a congestion control protocol that embodies the principle of decoupling congestion control from bandwidth allocation and uses this decoupling to achieve good performance in high bandwidth-delay networks. XCP provides a joint design of the end systems and the router and is characterized by the existence of separate congestion/efficiency controller and fairness controller at the router. XCP outperforms TCP both in traditional environments and in high bandwidth-delay product networks. Further, XCP maintains TCP’s scalability and robustness, and can provide a variety of qualities of service. We compared the performance, stability, and feasibility of XCP with other protocol and concluded XCP is an excellent protocol which provided with good performance, reasonable design, and feasibility. We considered it as the powerful candidate of the future high bandwidth-delay networks and choosed it to research one step more. Our research work include two aspects mainly: 1. Firstly, adding the admission control at the XCP receiver. High bandwidth-delay networks are environments where the per-flow bandwidth can reach a few Gbit/s, then the receiver’s speed should be taken into account. Aiming at this destination, we adjust the structure of the receiver system, adding the admission control. The receiver can modify the Feedback field to express the allowable speed. Then the field Feedback can reflect the whole network status, including the sender, the link on the path and the receiver. After such improvement, partial of traffic control function integrated into congestion control. It make traffic control more simple andclearly and cancel the field window in traffic control.2.Secondly, based on the research of the parameter of XCP, we discover that parameterinfecting the network utilization obviously, parameterinfecting the time of empting queuein the router, parameterinfecting the time of converging to system fairness status. All thethree parameter are the trade-off between of efficiency and stability. The utilization ofbackbone is not very high normally in fact. Parameterinfecting the network utilizationobviously, so through adjusting parameterin different situation to advanced theperformance of network. When the system is more stable and the utilization is not very high,we adjust the parameterhigher, contrarily, we adjust the parameterlower. Weuse the average queue to judge the system’s stability. Experiments prove thatimprovements make system more efficiency and maintain the stability.we use software Net Simulator-2 to complete experiments in this paper and use control theoryto analyzing the system stability.更多还原