《高速网络拥塞控制研究硕士毕业论文》由会员分享，可在线阅读，更多相关《高速网络拥塞控制研究硕士毕业论文（56页珍藏版）》请在金锄头文库上搜索。
1、 硕士学位论文 论文题目 高速网络拥塞控制研究 英文题目 Research on Congestion Control for High-Speed Network 硕士研究生 指导教师 学科专业 计算机软件与理论 论文提交日期 2009年4月 论文答辩日期 论文评阅人 答辩委员会主席 2009 年 4 月 10 日Abstract摘 要随着高速网络应用的日益广泛，拥塞控制机制的研究变得越来越重要。拥塞控制至少应该包含两部分：首先是要有源端算法响应路径中的拥塞，动态的调节数据发送速率;另一方面，要有一个中间节点管理机制能有效地预测、监测路径中的拥塞程度，通过显式或隐式的方法在拥塞发生前及时警告
2、源端。目前研究适合高速网络的TCP拥塞控制机制成为一个新的研究热点，一些研究者提出了一些新的算法如：STCP，H-TCP等。这些协议都是通过修改发送窗口的增加减小模式来提高TCP在高速网络中的性能。其中TCPW是以可用带宽测量为基础的新的TCP协议，对原有TCP协议改动较小，具有较好RTT公平性和较好的TCP友好性，在真实网络中易于实现，但是TCPW仍存在一些性能缺陷。由于TCPW窗口增长仍采用线性增加模式，因此不能像其他协议一样快速获得更大的发送窗口，而且在该算法的慢启动阶段仍然采用指数增长模式，从而导致大量突发数据的产生，造成拥塞。中间节点控制由路由器拥塞控制算法来实现，主动式队列管理机制
3、（AQM）是IEFT推荐的基于路由器拥塞控制关键技术，它和TCP端到端的拥塞控制相结合，是解决目前网络拥塞控制问题的一个主要手段。RED算法是AQM的一个典型，但其在算法稳定性和参数敏感性方面存在缺陷。本文基于以上两个算法，开展了以下三个方面的工作。首先对TCPW算法进行改进，主要集中在以下两点：一是在慢启动阶段发送窗口较原有算法能较快的到达10个包左右，之后窗口增长速度较原有算法有所减慢，这样有利于短流传输和避免突发数据产生，从而减缓拥塞；二是在拥塞避免阶段采用基于当前拥塞窗口大小的先快后慢的非线性增长方式，使之更适合于高速环境。通过建立新算法的数学模型分析其稳定性、RTT公平性和对TCP友
4、好性，在此基础上分别对以上两点改进采用NS2仿真方法加以验证，发现算法较原有算法在高速环境下有更好的吞吐量和更有利于短流数据传输。另外本文在分析RED算法基础上，提出了一种新的改进型AQM算法DRED算法。DRED相对RED算法，能够动态调整参数，并且采用非线性函数代替原有的丢包率计算方法。通过动态调整来调整向源端发送拥塞通知的速率，维持队列的稳定；通过新丢包率计算方式，提高缓冲的利用率和使队列长度尽量稳定于期望值附近。最后通过仿真来验证新算法更适应网络流量的变化，保持队列长度的稳定和丢包率的稳定，从而提高了网络链路利用率。关键词：高速网； 拥塞控制； TCPW； RED 47Abstract
5、AbstractWith the development of the applications on high-speed network, research on congestion control becomes more and more important. Congestion control should include two parts: end-to-end control and link control. End-to-end control could adjust the data sending rate dynamically in order to resp
6、ond to link congestion. On the other hand, the link control can predicate and monitor the degree of congestion effectively, then send the warning to sender before congestion happening by explicit or implicit method. Now more and more scientists begin on the researches of the TCP congestion control m
7、echanisms for high-speed networks and this research has become a focal subject. There are some typical protocols:, such as STCP, H-TCP,TCPW etc. These new protocols enhance the performance on the high-speed networks through adjusting the increases and decreases mode of window, TCPW algorithm is a be
8、tter one, it is based on BWE (Bandwidth Estimation) and has little changes on TCP protocol. it also provides a sensible fairness increment with respect to TCP Reno, Moreover, TCPW is friendly to Reno. But there are still some shortages in it, Congestion window of TCPW is based on additive increase o
9、f linear mode. So, in high-speed networks it cant obtain high throughput rapidly. During the slow-start stage, the congestion window of TCPW is based on exponential growth mode, then this will cause the datagram increases too fast and prompt the probability of congestion. Link control is implemented
10、 by router. The active queue management mechanism(AQM) is, which the IETF recommends, the essential technology based on the router congestion control, combining with the TCP end-to-end congestion control, it provide an effective method to solve the congestion control question on Internet. RED algori
11、thm is typical algorithm in AQM, but it exist some drawbacks, for example, the instability and the parameter sensitivity. In this paper, we give the researches on the following three aspects based on the above algorithm: TCPW and RED. At first, we improve the performance of the TCPW from two points.
12、 One enhances the former method by reducing the increasing speed of the datagram and increasing the increasing speed before the window is 10 during the slow-start stage. This decreases the occurring rate of congestion and improves the performance of short-term linkages transmission. On the other han
13、d, we use a simple nonlinear mode to increase the congestion widow instead of the linear mode. This new mathematical model and the new algorithm is friendly to Reno and have fairness increment with respect to TCP Reno. Through the simulation on NS-2 platform, the experiments show that the new algori
14、thm can obtain more throughput than TCPW in high-speed network and improve the short-term linkages transmission. Secondly, the other work an improved algorithm DRED of active queue management (AQM), is proposed. Based on the interpolations size, DRED can adjust dynamically the size Pmax, and therefo
15、re adjust the sending rate of congestion notification to the source end in time. The new algorithm uses the nonlinear mode to adjust the probability of drop, and maintain the stability of queue length of mathematical expectation. At last, the simulations on NS-2 show that DRED can adapt the change of network flow validly, hold the stability of queue length and also decrease the probability of drop. So it is superior to the RED algorithm in
