基于Linux的FAST TCP拥塞控制算法优化

发布时间：2018-10-14 17:41

【摘要】：现代网络及其应用取得了的令人瞩目的发展,使得链路承载了越来越多的数据,网络中传输数据的速度也要求更加迅速,现代的网络传输已不能满足更多数据、更快传输的要求。在未来,大数据、云计算、虚拟现实和增强现实等应用,以及广域互联、建立全联接世界的目标,呼唤新的技术使得网络能够传输更多的数据量,能够更快地传输数据。现在的网络,特别是广域网链路,带宽仍然比较小而时延和丢包率较大,传统的经典TCP传输算法已经无法满足现代以及未来大带宽长延时网络的需求,新的传输协议以及各种优化算法不断地被提出,但是因为或多或少的缺点,都无法广泛地使用。因此,本文选取了当前最新的TCP拥塞控制算法——FAST TCP[1],来探究其在当前广域网环境下的传输效果,研究其存在的不足,对存在的问题的进行改进。首先,本文详细地介绍了当前广域网的优化技术与方法,研究了 TCP拥塞控制机制,TCP吞吐量与链路带宽、时延以及丢包率的关系;其次搭建网络链路和模拟链路带宽时延,然后基于Linux平台对FASTTCP拥塞控制算法进行了研究,分析其存在的拥塞窗口波动性变化问题以及RTT公平性的问题,提出了平滑窗口波动的改进方法,并通过Linux内核提供的拥塞控制钩子,实现了改进的拥塞控制方案,最后将FASTTCP算法优化前后对比测试,验证了该优化能有效地平滑窗口的波动并提高了两条流的RTT公平性。本文的主要工作如下:基于Linux TC工具模拟广域网带宽、延时、丢包等参数来搭建链路;基于Linux平台,创建动态内核模块,实现FASTTCP拥塞控制算法;提出并实现了该改进的FASTTCP拥塞控制算法,然后对比测试了算法优化前后以及目前广泛使用的CUBIC算法之间的传输效果。
[Abstract]:With the remarkable development of modern network and its application, the link is carrying more and more data, the speed of data transmission in the network is also required to be faster, and the modern network transmission can no longer meet more data. Faster transmission requirements. In the future, big data, cloud computing, virtual reality, augmented reality and other applications, as well as wide-area interconnection, the goal of establishing a fully connected world, calls for new technologies to enable the network to transmit more data and data faster. In current networks, especially in WAN links, the bandwidth is still relatively small and the delay and packet loss rate is large. The traditional classical TCP transmission algorithms can not meet the needs of modern and future large-bandwidth long-delay networks. New transport protocols and various optimization algorithms have been proposed, but because of more or less shortcomings, they can not be widely used. Therefore, this paper selects the latest TCP congestion control algorithm, FAST TCP [1], to explore its transmission effect under the current WAN environment, to study its shortcomings, and to improve the existing problems. First of all, this paper introduces the current WAN optimization techniques and methods in detail, studies the TCP congestion control mechanism, the relationship between TCP throughput and link bandwidth, delay and packet loss rate, and then builds network link and analog link bandwidth delay. Then, based on the Linux platform, the congestion control algorithm of FASTTCP is studied. The problem of fluctuation of congestion window and the fairness of RTT are analyzed, and an improved method of smoothing window fluctuation is proposed. An improved congestion control scheme is implemented through the congestion control hook provided by the Linux kernel. Finally, the FASTTCP algorithm is tested before and after optimization, which verifies that the optimization can effectively smooth the fluctuation of the window and improve the RTT fairness of the two streams. The main work of this paper is as follows: simulating WAN bandwidth, delay, packet loss and other parameters to build link based on Linux TC tools, creating dynamic kernel module based on Linux platform, realizing FASTTCP congestion control algorithm; The improved FASTTCP congestion control algorithm is proposed and implemented. Then, the transmission effect of the improved FASTTCP congestion control algorithm before and after the optimization and the widely used CUBIC algorithm is compared and tested.
【学位授予单位】：南京大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP393.0

【相似文献】