【作者】 董林芳；

【导师】 舒炎泰；

【作者基本信息】 天津大学 ， 计算机应用技术， 2007， 博士

【摘要】 随着无线网络越来越广泛地应用于我们的工作和生活中,如何在保证效率的同时提高TCP公平性成为一个重要的研究课题。本论文主要研究多跳无线Ad Hoc网络中TCP业务的公平性问题,分析了造成TCP业务不公平性的多种原因,推导出802.11 MAC层的多跳延迟,在此基础上提出基于MAC层传输延迟的邻居区域随机提前检测机制(D_NRED)。另外,本文提出自适应暂停机制(Adaptive_Pause)来动态分布式实现提高TCP公平性。无线网络中传输的业务有些对延迟是非常敏感的,而且延迟也是一个很重要的公平性指标。本文对IEEE 802.11 MAC层的延迟进行分析,首先分析不饱和业务下的单跳平均延迟,进而扩展到多跳情况下进行分析,得到端到端的平均延迟。分析过程中引入了二进制指数退避机制的二维Markov模型和M/G/1排队模型。通过大量的仿真对分析结果进行了验证,结果显示模型求解的分析值与仿真结果拟合的很好。MAC层延迟与信道竞争的激烈程度之间存在着定性的关系,MAC层的介质访问延迟能够反映出无线信道的拥塞程度。本文提出基于MAC层传输延迟的D_NRED机制来检测拥塞、调节丢包率并提高TCP公平性。每个节点测量数据包从离开MAC层队列到成功收到ACK的时间延迟,以此作为判断邻域拥塞情况的标准,调节丢包概率,达到提高公平性的目的。该机制主要包括三部分内容:判断拥塞发生在哪里;发出拥塞通知;节点计算丢包概率。本方法测量MAC层传输延迟不需要修改硬件,通过修改驱动程序就可以在实际应用中实现,具有更好的可行性。仿真结果表明该机制可以提高无线网络中TCP流的公平性。Adaptive_Pause机制是一种用以提高TCP公平性的动态分布式控制机制。节点根据竞争信道的邻居节点数目计算自己应该占有的无线信道带宽份额。如果节点占用的带宽超过了公平性份额,就在一段时间内主动地暂停发包,使得其他节点有机会发送数据。该机制的特点是:对网络中的不公平情形反应迅速,在信道竞争中处于有利地位的节点主动让出信道,而不是等受压制的节点发出通知才开始做出反应。我们利用二维Markov模型分析带有pause状态的MAC层指数退避机制,给出最优pause时间和带宽比例阈值。通过仿真验证了理论推导的正确性和Adaptive_Pause机制对于提高TCP公平性的效果。更多还原

【Abstract】 As wireless networks are becoming increasingly prevalent, the problem of ensuring TCP fairness is an increasingly important issue. In this dissertation, we investigate the problem of TCP fairness in multi-hop wireless Ad Hoc networks. The causes of TCP unfairness are studied and the multi-hop MAC delay is analyzed. We propose a MAC delay based Neighborhood RED scheme (D_NRED) to improve the TCP fairness. Moreover, a dynamic distributed scheme—Adaptive_Pause is proposed.Much traffic in Wireless Ad Hoc networks is very sensitive to delay, and delay is an important metric of fairness. We analyze the average packet delay on IEEE 802.11 DCF under finite load traffic in multi-hop Ad Hoc networks. We employ a Markov chain model to analyze the channel access delay. We model each node using an M/G/1 queue and derive the queueing delay. Thus the average packet delay in single-hop scenario can be obtained. The model is also extended to evaluate the end-to-end packet delay in multi-hop Ad Hoc networks. To validate our analytic results, we have done extensive simulation. The results from analysis and simulation match closely.As it is known, the packet delay is large if the wireless channel is very busy and the overall traffic load exceeds the capacity of the channel. Thus the packet delay can reflect whether or not the channel is busy. In this dissatation, we propose D_NRED scheme to detect channel congestion, tune the dropping probability, and improve TCP fairness. The scheme detects congestion according to MAC delay and notifies the nodes which use too much channel capacity to drop their packets. Thus the expressed node can get a chance to transmit. This scheme includes three parts: how to find out the congestion center correctly and effectively; how to inform its neighbors about the congestion; and how the neighbors to calculate their local packet drop probability. The MAC delay can be measured by modifies the driver instead of the hardware.Therefore, D_NRED scheme has a good practical implement. Simulation results show that D_NRED scheme can improve the TCP fairness.We propose a simple but efficient fairness mechanism: Adaptive_Pause, which is distributed and low cost at implementation. Each node monitors the occupation of the channel due to its emissions and determines whether it should pause an interval in order to avoid channel capture. The characteristic of the scheme is that the nodes which consume too much bandwidth share will actively pause their emission. Comparing to the passive scheme in which the node controlling their transmission when received a message packet from the node at the expressed location, the proposed scheme is more promptly and with less overhead. By using a two dimensional Markov model, the optimal parameters are found. These parameters can be used by the nodes to dynamically determine their fair share. Simulation results validate the correction of our analysis and the scheme’s effect in TCP fairness improvement.更多还原