AQM中参数依赖性的研究与改进
发布时间：2018-08-17 13:47
【摘要】：随着经济的不断发展，网络上的负载也越来越重，越来越多的网络拥塞现象开始显现。所以，对于拥塞现象控制的研究也越来越重要。拥塞控制的机制根据其实现位置，可分为两大类：基于源端的算法（Source Algorithm）和基于链路的算法（Link Algorithm）。基于源端的算法主要以TCP协议内容中拥塞控制机制的部分为主。虽然TCP协议中的拥塞控制机制在网络中已成功运行了很多年,但是仅仅依靠TCP协议提供的拥塞控制机制是无法满足不断增长的网络数据需求的。这就要求我们要将眼光放到网络中间节点上去。现今，基于链路算法的拥塞控制机制主要集中于中间节点的缓存队列管理机制上，主要分为被动队列管理机制(PQM，Passive Queue Management)与主动队列管理机制(AQM, Active Queue Management)。由于被动队列管理存在“死锁”、“全局同步”等问题，而且主动队列管理可以预见地而非反应式地丢弃分组，因而主动队列管理机制得到了更多研究者的关注。其中早期随机检测(Random Early Drop, RED)算法是主动队列机制中典型的应用算法。 本文对主动队列管理算法进行了详细研究，其中包括RED、ARED、GRED、FRED等算法。针对RED算法中参数敏感性的问题，，对RED算法进行了改进。并且利用网络仿真工具在模拟的网络拥塞环境下对算法性能进行分析。本文设计了三种基于RED的改进算法，RED-S，RED-T及RED-P。其中RED-P算法利用二次函数特性来增加丢包概率变化的平滑性，减小了平均队列长度的变化率，并且在不同区间采用了不同公式计算丢包概率，增加拥塞控制力度的同时保证了吞吐量，让平均队列长度可以平滑的过渡，性能也最优。本文选用了NS2这一网络仿真工具，对改进前后算法的运行进行了仿真，对拥塞场景进行了设置，并且对改进协议前后端到端的平均延时，平均队列的长度以及吞吐量等参数进行了比较。 仿真表明：本文提出的RED-T，RED-S，RED-P算法能有效的改进RED的参数敏感性的问题，适度简化了参数设置。并且，在减小瓶颈段端到端延时的同时减小了平均队列的波动范围与抖动幅度，能维持几乎接近的吞吐量。在保证改进算法服务质量的前提下，可以更好的应付网络的拥塞。
[Abstract]:With the development of economy, the load on the network becomes more and more heavy, and more network congestion appears. Therefore, the study of congestion control is becoming more and more important. Congestion control mechanisms can be divided into two categories according to their location: source-based algorithm (Source Algorithm) and link-based algorithm (Link Algorithm). The source-based algorithm is mainly based on the congestion control mechanism in the content of TCP protocol. Although the congestion control mechanism in the TCP protocol has been running successfully in the network for many years, the congestion control mechanism provided by the TCP protocol alone can not meet the increasing demand for network data. This requires us to look at the network in the middle of the node to go. Nowadays, the congestion control mechanism based on link algorithm is mainly focused on the buffer queue management mechanism of intermediate nodes, which is divided into passive queue management mechanism (PQMN passive Queue Management) and active queue management mechanism (AQM, Active Queue Management). Because of the problems of "deadlock" and "global synchronization" in passive queue management, and active queue management can discard packets predictably rather than reactive, active queue management mechanism has attracted more attention. Among them, early random detection (Random Early Drop, RED) algorithm is a typical application algorithm in active queue mechanism. In this paper, active queue management algorithms are studied in detail, including red / red algorithm. Aiming at the problem of parameter sensitivity in RED algorithm, the RED algorithm is improved. Network simulation tools are used to analyze the performance of the algorithm in the simulated network congestion environment. Three improved algorithms based on RED are designed in this paper. The RED-P algorithm makes use of quadratic function to increase the smoothness of packet loss probability, reduces the change rate of average queue length, and uses different formulas to calculate packet loss probability in different regions. Increase congestion control while ensuring throughput, so that the average queue length can smooth the transition, and the performance is optimal. In this paper, the network simulation tool NS2 is used to simulate the algorithm before and after the improvement, and the congestion scenario is set up, and the average end-to-end delay before and after the improved protocol is given. The average queue length and throughput are compared. Simulation results show that the proposed RED-TX RED-S RED-P algorithm can effectively improve the parameter sensitivity of RED and simplify the parameter setting appropriately. In addition, the end-to-end delay of the bottleneck segment is reduced while the fluctuation range and the jitter amplitude of the average queue are reduced, which can maintain an almost close throughput. Under the premise of improving the QoS of the algorithm, it can better deal with the congestion of the network.
【学位授予单位】：内蒙古科技大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP393.06
本文编号：2187834
[Abstract]:With the development of economy, the load on the network becomes more and more heavy, and more network congestion appears. Therefore, the study of congestion control is becoming more and more important. Congestion control mechanisms can be divided into two categories according to their location: source-based algorithm (Source Algorithm) and link-based algorithm (Link Algorithm). The source-based algorithm is mainly based on the congestion control mechanism in the content of TCP protocol. Although the congestion control mechanism in the TCP protocol has been running successfully in the network for many years, the congestion control mechanism provided by the TCP protocol alone can not meet the increasing demand for network data. This requires us to look at the network in the middle of the node to go. Nowadays, the congestion control mechanism based on link algorithm is mainly focused on the buffer queue management mechanism of intermediate nodes, which is divided into passive queue management mechanism (PQMN passive Queue Management) and active queue management mechanism (AQM, Active Queue Management). Because of the problems of "deadlock" and "global synchronization" in passive queue management, and active queue management can discard packets predictably rather than reactive, active queue management mechanism has attracted more attention. Among them, early random detection (Random Early Drop, RED) algorithm is a typical application algorithm in active queue mechanism. In this paper, active queue management algorithms are studied in detail, including red / red algorithm. Aiming at the problem of parameter sensitivity in RED algorithm, the RED algorithm is improved. Network simulation tools are used to analyze the performance of the algorithm in the simulated network congestion environment. Three improved algorithms based on RED are designed in this paper. The RED-P algorithm makes use of quadratic function to increase the smoothness of packet loss probability, reduces the change rate of average queue length, and uses different formulas to calculate packet loss probability in different regions. Increase congestion control while ensuring throughput, so that the average queue length can smooth the transition, and the performance is optimal. In this paper, the network simulation tool NS2 is used to simulate the algorithm before and after the improvement, and the congestion scenario is set up, and the average end-to-end delay before and after the improved protocol is given. The average queue length and throughput are compared. Simulation results show that the proposed RED-TX RED-S RED-P algorithm can effectively improve the parameter sensitivity of RED and simplify the parameter setting appropriately. In addition, the end-to-end delay of the bottleneck segment is reduced while the fluctuation range and the jitter amplitude of the average queue are reduced, which can maintain an almost close throughput. Under the premise of improving the QoS of the algorithm, it can better deal with the congestion of the network.
【学位授予单位】：内蒙古科技大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP393.06
【参考文献】
相关期刊论文 前10条
1 杨晓萍,陈虹,翟双;基于路由器的RED和Droptail算法比较[J];吉林大学学报(信息科学版);2005年01期
2 吴春明,姜明,朱淼良;几种主动式队列管理算法的比较研究[J];电子学报;2004年03期
3 张振平;;一种基于平均队列长度的改进RED算法[J];计算机光盘软件与应用;2012年12期
4 田立伟;陈元琰;罗晓曙;;一种新的TCP/AQM模型的高带宽拥塞控制算法[J];计算机工程与应用;2009年30期
5 张鹤颖;窦文华;;主动队列管理机制的性能分析[J];计算机科学;2004年04期
6 武航星;慕德俊;龚贤武;潘文平;;FERED:公平性增强的RED算法[J];计算机科学;2009年02期
7 王建新;荣亮;肖雪峰;;几种主动队列管理算法的仿真及性能评估[J];计算机工程;2007年03期
8 张鹤颖,窦文华;主动队列管理机制在突发流下的性能比较与分析[J];计算机工程与科学;2004年12期
9 任丰原,林闯,刘卫东;IP网络中的拥塞控制[J];计算机学报;2003年09期
10 周朋;惠晓威;;自适应拥塞控制算法的研究与仿真[J];计算机仿真;2010年09期
本文编号：2187834
本文链接：https://www.wllwen.com/guanlilunwen/ydhl/2187834.html
