【作者】 曹振臻；

【导师】 肖扬；

【作者基本信息】 北京交通大学 ， 信号与信息处理， 2010， 博士

【摘要】 本文围绕用户通过无线局域网与Internet中的服务器通信中若干关键问题进行模型和算法研究,贡献及创新点包括以下几个方面：1802.11无线局域网MAC层的有限负载模型。首先,用户通过无线局域网与Internet服务器相连,无线局域网的性能直接关系到用户与服务器通信的性能。无线局域网的MAC层协议对于无线局域网的性能发挥着重要作用。我们提出了一种有限负载条件下802.11 MAC层协议的数学模型,该模型利用二维Markov链分析了802.11无线局域网节点的退避过程及空闲等待过程,通过求解Markov模型的稳态概率,计算出802.11无线局域网MAC的吞吐量,并得出802.11 WLAN MAC层接入延时。我们根据模型结果,首次提出了802.11MAC在饱和与非饱和临界条件下的不稳定问题,并通过仿真验证了分析结果。2多跳无线局域网的MAC层优化协议。当前的单跳无线局域网存在着覆盖范围小等问题,目前IEEE 802.11s工作组正在制定一种多跳的Mesh WLAN标准,因此多跳的802.11无线局域网是无线局域网研究的前沿课题。本文分析了多跳IEEE 802.11网络的节点传输失败原因的非对称性(即节点在不同传输方向上的导致传输失败原因不相同的特性),针对已有的802.11 MAC采用相同的失败重传机制存在的问题,提出在多跳IEEE 802.11 MAC协议中对于不同的传输方向采用非对称的退避机制,进而提出了MAC层优化协议DCF-CD。DCF-CD主要思想是：根据对节点传输失败概率和退避过程中时隙繁忙概率的估计,在节点进行传输失败重传时以计算得到的概率增大退避时间,通过提高多跳802.11网络中信道空间复用能力,来提高网络的整体吞吐能力。本文提出的优化协议不需要对物理层提供额外要求,降低了该优化协议实现的复杂度。3 Internet路由器主动队列管理模型与算法研究。用户与服务器在Internet上通信的关键问题是Internet拥塞控制,路由器主动队列管理是Internet拥塞控制的重要机制。本文中,我们根据TCP/AQM流体模型的分析提出了TCP/AQM系统稳定的充分条件。流体模型存在着不足,我们利用2-D Markov链提出了一种新的TCP/AQM时延闭环反馈模型,它不仅能有效地分析TCP发送窗口、路由器队列长度(排队时延)及路由器丢包率的平均值,还能通过Markov链的求解得到TCP/AQM系统重要参数的稳态分布。对于TCP-UDP混合流,本文研究了路由器TCP与UDP区分服务的队列机制,并将TCP/AQM Markov模型扩展为TCP-UDP/AQM模型,以分析UDP流对TCP/AQM系统的影响；同时我们提出了一种在单一队列中区分不同优先级UDP流的区分服务的主动队列管理算法,它能够通过对队列状态的估计调整丢包策略以减少UDP流的丢包率。4用户行为与流量模型和服务器参数优化算法。影响用户与服务器通信性能的另一关键问题是服务器拥塞控制。我们提出了用户行为与服务器消息流量的数学模型。我们根据对用户随时间行为变化的分析,提出了一种用户状态转移模型,根据用户行为与服务器消息流量的关系以及系统消息相互之间的关系,提出了服务器的流量模型。根据该模型可以估计到达服务器的消息流量,以用于设计服务器容量以及预测服务器潜在的拥塞风险。此外,我们分析了服务器中一种特殊消息队列QCVBA(集束式到达服务器带休假队列)的模型,并根据模型分析的结果提出了一种服务器的参数的优化算法。利用该算法,服务器能够在满足消息队列的QoS要求的前提下,获得最长的休假时间处理其它事务,从而提高服务器的性能。更多还原

【Abstract】 This thesis studies on the models and schemes of some key issues in the communication between users and server via Wireless Local Area Network (WLAN) and Internet. Its contribution is as follows:1. The model of 802.11 MAC layer in finite load condition. When users access to Internet server via WLAN, the performance of WLAN directly influences the performance of users-server communication. The MAC layer protocol plays a crucial role in WLAN. We propose a novel model of 802.11 MAC in finite load condition. This model implements a 2-D Markov chain to analyze the backoff periods and waiting periods of nodes in 802.11 WLAN, and the stationary distribution of the Markov chain is calculated. By this model, not only the system throughput but also the access delay in MAC layer can be obtained. The instability problem of 802.11 WLAN in the saturated-unsaturated border field is firstly proposed based on this model, and this model is validated by simulations.2. The enhanced MAC protocol for multihop WLAN. There are some shortages in the single-hop WLAN currently used, such as limited coverage. IEEE 802.11s Task Group is developing a standard for multihop Mesh WLAN, so it is an advanced topic to study on the multihop 802.11 WLAN MAC. This thesis analyzes the asymmetry of transmission failures in different directions (which means in different directions a node may encountes different kinds of failures) and points out the reasons why current 802.11 MAC which uses symmetrical backoff mechanism operates inefficiently in the multihop WLAN. We propose an enhanced MAC protocol called DCF-CD to use asymmetrical backoff mechanisms in different directions. The basic idea of DCF-CD is:based on the estimation of transmission failure probability and slot busy probability in backoff period, when a node transmits unsuccessfully it increases its backoff period with a calculated probability. This protocol can improve the channel spatial reuse in multihop 802.11 WLAN and increase the total throughput. The proposed protocol does not require any additional overheads in physical layer, so it is easy to be implemented.3. Research on model and scheme of Active Queue Management (AQM) in Internet router. Congestion control is a key issue when users communicate with server via Internet. AQM is an important Internet congestion control mechanism. In this thesis, based on the fluid model we propose a sufficient condition for stability of TCP/AQM system. For some shortages in fluid model, we propose a novel model of TCP/AQM delayed closed-loop feedback system based on 2-D Markov Chain. This model can estimate the mean values of TCP congestion window, router queue length (queuing delay) and packets dropped probability. The distribution of some key parameters can also be obtained from the equilibrium distribution of the Markov chain. For TCP-UDP mixed flows, we study on the multi-queues mechanism, and extend the TCP/AQM Markov model to a TCP-UDP/AQM model to analyze the TCP/AQM system with UDP flows existing. For the UDP queue, we propose a differentiated services (DiffServ) AQM scheme for multi-priorities UDP flows, which can decrease the UDP packets dropped ratio by estimating the queue states.4. The user behavior and traffic model and a scheme to optimize server’s parameters. In addition to the issues in WLAN and Internet, congestion control in server is another key issue affecting the performance of users-server communication. We propose a model analyzing the user behavior to estimate the messages traffic in server. Firstly, we analyze the user behavior in a day, and then propose a user behavior model, and based on the relationship between user behavior and system traffic and the relationship among messages, we propose a model of the messages traffic in server. Moreover, we analyze a special messages queue QCVBA (Queue with Controlled Vacation and Batched Arrival) in server, and based on analysis results we propose a scheme to optimize the server parameters. By the proposed scheme, when handling QCVBA queue, the server can obtain optimal vacation time with satifying the basic QoS requirements, to make server performance improved.更多还原