【作者】 郭永安；

【导师】 陈春玲；

【作者基本信息】 南京邮电大学 ， 计算机软件与理论， 2012， 硕士

【摘要】 广域网文件传输速度慢是目前各大网络公司面临的问题,针对这些情况,探讨了采用重复数据消除技术减少数据传输量,从而提高网络传输速度。在实现重复数据消除的网络节点,探讨了队列管理、重复数据消除技术、排队论等相关技术的基本概念及研究现状,提出了基于TCP透明代理的队列管理模型。针对网络节点产生的队列A和队列B建立了两个排队模型,模型一为单服务台排队模型,即队列A和队列B共享给定的内存空间;模型二为两级串联排队网络模型,即队列A和队列B平分给定的内存空间。在理论上推导出模型一和模型二在零丢包率时所需的内存空间。根据网络节点用户连接数与内存需求的正比关系,建立并实现了动态队列管理模型,其核心是在网络节点中根据用户连接数的大小动态调整队列管理方式,使网络丢包率、队列长度和排队时延达到平衡。在实际研发过程中,构建了广域网TCP透明代理平台,分别调整给定的队列空间,利用模型一和模型二进行验证。实验结果表明,在内存空间较大的情况下,模型二优于模型一,但在内存空间不够的情况下,模型一优于模型二。将排队论的相关理论用于网络节点中队列管理中,能够在内存有限同时实际用户连接数不断变化时,采用动态队列管理模型,使丢包率、队列长度和排队时延达到平衡。更多还原

【Abstract】 The low speed of transmission in Wide Area Network (WAN) is a problem that every internet company wants to solve currently. According to the situation, the paper study on reducing the amount of data transmission by eliminating duplicated data to improve the network transmission speed.It is in the network node which implements elimination duplicate data. The paper discuss the basic concepts and research status of queue management, data redundancy elimination, queuing theory and other related technology, propose the queue management model based on the TCP transparent proxy. It establish two queue models for queue A and B, one of the models is a single server queue model that queue A and B share the given memory space and another is a two stages tandem queue model that queue A and B bisect the given memory space, we name them model 1 and model 2.It derive the size of memory space needed that zero packet loss rate in theory. The establishment and the implementation of the dynamic queue management model is according to the proportional relationship of the number of user connections and memory requirements. Its core is dynamic adjust queue management mode according to the number of user connections size, so that the packet loss rate, queue length, queue delay keep balance. In the process of practical project development, construct the WAN TCP transparent proxy platform, verify the two models by adjusting the given space. The experimental results show that in the case of large memory space, model 2 is better than the model 1, but model 1 is superior to model 2 in the case of the memory space is not enough.Apply queuing theory to queue management in the network node can make the packet loss rate, queue length ,queue delay achieve a balance by dynamic adjust queue management models when the memory is limited and the actual number of user connections is constantly changing.更多还原