节点文献

网络拥塞控制及RED算法改进策略研究

Congestion Avoidance Control and RED Algorithm Modification Research

【作者】 苑红晓

【导师】 杨静;

【作者基本信息】 山东大学 , 计算机系统结构, 2007, 硕士

【摘要】 随着互联网规模和高速网络中多媒体应用的发展,人们对网络服务质量(QoS)的需求越来越高,不但对网络有很高的带宽要求,而且要求信息传输的低延迟和低抖动等,需要提供端到端的QoS控制和保证。而网络拥塞是影响网络服务质量的重要因素,实施拥塞控制也是其它QoS机制正常工作的必要前提。因此,如何避免拥塞、如何进行拥塞控制保证QoS是当前的研究热点。主动式队列管理机制(AQM)是IETF推荐的基于路由器拥塞控制的关键技术,它和TCP端到端的拥塞控制相结合,是解决目前Internet拥塞控制问题的一个主要途径。AQM通过评估网络状态、预测拥塞的出现,对分组进行有目的的丢弃,从而可以使发送端更及时地了解到网络状况并调整发送速率。RED(RandomEarly Detection)算法就是AQM的一个典型代表,但是现有算法在响应速度、稳定性及环境敏感性等方面仍有缺陷。对此,本文通过对的RED算法进行详细分析的基础上,总结出已有算法的优势和不足,提出了一种新的AQM算法——CARED(Cauchy Adaptive RED)算法。CARED算法对原有RED算法的分组丢弃概率Pb的计算方法和参数Pmax的取值进行修改。其一:利用模糊理论中的升半哥西分布的隶属函数代替原来的线性增加分组丢弃概率的函数。CARED分组丢弃概率计算采用升半哥西分布函数,以平均队列长度为样本来获得,将控制范围扩展为最小阈值到最大缓冲之间,实现了分组丢弃概率变化的平滑化,保证在每一阶段系统都能迅速地对拥塞作出反应。其二:CARED通过计算出路由器队列单位时间间隔内的平均队列长度,分别与最大阈值或最小阈值的比较,根据二者差值的大小动态地调整Pmax的大小,调整向源端发送拥塞通知的速率,维持队列长度的稳定,避免不必要的传输延时和抖动。并且Pmax的值不是每个时间间隔都更新,而只有当连续两个时间间隔内Status都处于相同状态,那么就可以更精确的认为缓冲队列中负载过大或过小,算法就会动态增加或减少Pmax的值,这样避免了Pmax的改变过于频繁,从而提高了链路的利用率。其三:CARED算法的健壮性来自它对Pmax的规律性调整,如果大量突发数据包导致网络拥塞程度发生急剧变化,则Pmax则需要过一段时间甚至10秒或20秒才能适应。为了保证算法在这段时间里性能不会过度下降,本文将Pmax的范围限制在[0.01,0.5]之间,这样,即使这段时间内平均队列长度Qav不在目标范围内,平均延时和吞吐量也不会下降太多,使算法虽然在理论使不能得到最优,但其性能可得到保证。在NS2网络仿真器上对算法进行了验证,一系列仿真实验表明,CARED能够有效地适应网络流量的变化,保持队列长度的稳定,减少了队列溢出和空闲现象的发生,在保持队列长度稳定以及提高链路利用率方面明显优于RED算法。

【Abstract】 With the unceasing development of Internet scale > people have an ever-growing demand for the quality of service of computer networks (QoS). And nowadays, in the high-speed network multimedia’s application not only has the very high bandwidth requirement to the network, but also requires intelligence transmission low delay and the low vibration and so on, needs to provide the endtoend control and guarantee of QoS. The network congestion is a great factor to affect the network quality of service , so the implementation congestion control is prerequisite for other QoS mechanism normal work .At present, congestion control mechanism of Internet and quality of service are the central issues of the current research.The active queue management mechanism( AQM) is, which the IETF recommends, the essential technology based on the router congestion control, which combines with the TCP end-to-end congestion control , being a main method to solve the congestion control question of the present Internet .By evaluation the state of network and foretelling the appearance of the congestion, AQM can drop the packet purposefully so that the sending end can be informed of the state of network and then adjust it’s sending rate. But the current algorithms aren’t still perfect in terms of response’s time, stability and sensitivity to the environment and so forth. In this paper, the advantages and disadvantages of the existent algorithms are concluded based on analysing the current prevalent congestion control algorithms RED in detail, an improved algorithm CARED of active queue management (AQM) is proposed.As a result, CARED algorithm has made the modifications of original RED algorithm in the computational method of probability of packet drop -Pb and the adapting of the parameter -Pmax. First, based on the fuzzy math, instead of the probability function of drop, we use the membership function of ascend demi-cauchy distribution. Original RED algorithm based on the average queue size of the buffer to adjust probability of drop, and when the average queue size reaches the Qmax(denoted by Qmax), the dropping rate increases linearly from Pmax to 1. This jump will aggravate the jitter of buffer queue. CARED improve probability calculation used a ascend half-cauchy distribution function, to extend the scope of control from the minimum threshold to the buffersize. It can also achieve the smoothness from part to complete marking or discarding. Second : CARED calculates the average queue length in the router unit- time-gap, and by comparing it with the maximum value and the minimum value, the interpolation can be obtained. Based on the interpolation’s size, CARED can adjust dynamically the size of Pmax, and therefore adjust the sending rate of congestion notification to the source end in time and maintain the stability of the queue length, in order to avoid the unnecessary delay of transmission and vibration .The algorithm is verified in NS2 network simulation machine as well. By the indication of a series of simulation experiments, CARED can validly adapt the change of network flow effectively, hold stability of queue length , reduce phenomenon occurrence of the queue overflow or the idle greatly. And It is superior to the RED algorithm in maintaining the stability of queue length and enhancing the utilization ratio of the links.

  • 【网络出版投稿人】 山东大学
  • 【网络出版年期】2007年 03期
  • 【分类号】TP393.07
  • 【被引频次】13
  • 【下载频次】444
节点文献中: 

本文链接的文献网络图示:

本文的引文网络