【作者】 李大双；

【导师】 周明天；

【作者基本信息】 电子科技大学 ， 计算机应用技术， 2002， 博士

【摘要】 通常，宽带网络一般指带宽至少为155Mb／s以上的网络，其带宽分配是以服务质量(QoS)控制为核心，包括网络拥塞控制、呼叫接纳控制(CAC)、QoS路控制以及VP带宽分态控制等方面。MPLS技术的出现，将ATM与IP技术各自的优点融合起来，加上资源预留协议(RSVP)以及区分服务(DiffServ)技术逐渐成熟，使得IP+ATM这种架构的Internet为多媒体业务流提供QoS保证成为可能。 本文围绕带宽分配的业务流拥挤控制技术、QoS选路技术、CAC技术、VP。带宽动态分配控制技术以及DiffServ带宽分配控制技术，在理论与实践上开展了深入的研究，并取得了若干创新性的技术成果： ●提出了一种基于碰撞——墙壁函数的业务流自适应分布的优化控制技术。在ATM网络中，用户业务流采用统计复用的方式来利用信道，业务流具有突发性，使业务流量的稳定性较差，波动幅度较大，很容易发生阻塞，而造成信元丢失，严重地影响服务质量。尽管已有多种ATM网络的流量控制技术，如呼叫连接接纳控制、QoS选路、业务流警管、选择性的信元丢弃、反应式拥塞控制和业务流成型等技术。但是，这些技术多是针对每个连接或某个业务流方向的局部性的拥塞控制技术，没有从整体上综合考虑全网的业务流分布情况来进行优化控制。论文提出的以信道带宽占用率为主要控制参数构成距离测度的碰撞——墙壁函数，对网络信道上的业务流拥挤很敏感，因而网络可以对业务流拥挤作出快速反应，避开业务流量拥挤的信道，实现业务流量分布的自适应优化控制，使新入网的业务按自适应算法得到的最优路由原则选路，可以使ATM网内的业务流量合理地、比较均匀地分布，避免网络中的各条信道上产生业务流量的拥塞。它是根据整个网络的业务流分布情况，从全局的角度去预防、减轻以至最终消除拥塞。 ● 提出了一种基于蚂蚁网络的新的QoS选路技术。在研究了蚂蚁网络选路机制的基础上，提出了一种探测网络链路状态变化、支持QOS 选路的新算法和实现路由信息分布的技术，即由具有较大带宽需求的呼叫驱动人工蚂蚁的发射，这样既能实时反映网络链路较大的变化情况，又能大大地减少人工蚂蚁的数量，降低网络链路对高呼叫到达率的处理负担。t 提出了一种采用神经网络和屏蔽遗传算法动态分配VP带宽的新算法。目前，虚路径（VP）已成为宽带网络中的关键技术。论文根据多速业务流模型，提出了采用神经网络与屏蔽遗传算法的一种控制算法，来为宽带网络的虚路径分配带宽。该算法能根据各类业务流的到达、服务特性和网络环境的情况，由神经网络来估计VP 子网的利用率，并应用遗传算法，自适应地为各VP 子网分配最佳带宽。由于它是一个带约束的优化问题，传统的遗传算法对它不再适用。因此，论文提出应用屏蔽遗传算法来解诀该优化问题。＠提出了一种基于测量和神经网络估计的接纳控制新算法。由于AT M网络中业务流特性以及Q O S要求的多样性，由分析模型精确估计统计复用业务流的QOS十分困难。论文提出一种基于复用业务流到达率的滑动窗日测量和神经网络估计的新方法，能够根据业务流的多个特征参数，快速作出 QOS估计，实现实时的接纳控制。这种方法能在动态环境中充分利用业务流的统计复用的特性，满足不同业务种类的QOS要求。．设计了基于虚拟定时以及 ATM捷径为 Diffsery提供 QoS的带宽分配控制技术。Diffsery是解决宽带 Internet QoS 控制的一种可行方案，它在边界节点将不同的QOS要求映射成PHB。论文对每类具有相同PHB的 P业务流，应用VTRR公平调度算法，在竟争带宽的业务流中实现带宽公平共享分配。将VTRR 与支持建立ATM捷径的RSVP结合使用，能为Diffsery提供QoS保证。更多还原

【Abstract】 In general, the bandwidth is at least 155Mbps in broadband networks, where bandwidth allocation is aimed at the control for Quality of Service(QoS), including network congestion control, Call Admission Control, QoS routing algorithm , virtual path’s bandwidth allocation, and so on. The technologic advantages of ATM and IP are merged together by MPLS or InterServ. Using the RSVP and DiffServ technology in IP plus ATM framework, it is possible to provide QoS guarantee for multimedia session.This dissertation has conducted researches on these technologies around bandwidth allocation and QoS control. The main contributions of this dissertation include:·Proposing a new adaptive control algorithm for optimally distributing traffics based on collision function of distance metric.As the channel is used on statistical multiplexing mode in ATM networks, the session’s traffics are generated burstly, it will lead the multiplexed flow poor in stability, so the aggregated traffic will fluctuate drastically. The more serious the congestion status are, the more the cells are discarded because of the block and overrun, it is difficult to guarantee the QoS. The existed traffic control technologies are algorithms based on each session only, so the global congestion control is needed to solve this problem, we provide a new method based on collision function sensitive to congestion, and use the collision function as the distance measurement of transmission channel. With this method, the network can response to congestion in local regions quickly and the routes in congesting status can be avoided, so as the adaptive control of optimal traffic distributing can be achieved, and thecongestion can be prevented in advance. Once congestion is appearing at some part in networks, it can be eliminated gradually by this algorithm.· Presenting a new QoS routing algorithm based AntNet. Weprovide a new QoS routing algorithm that is based on AntNet but different from the existed AntNet algorithms. The changes of link state can be detected in time by this algorithm, it can support global QoS routing perfectly. The emit of artificial ant within this method is driven by calling with the higher bandwidth requirement, the number of artificial ant is reduced largely, and the processing load is cut down when the request arrival rate is relatively high· Proposing a new method for bandwidth allocation of virtual paths using neural networks and masked genetic algorithms based on the model of multirate’s traffic. The concept of virtual paths has become the key technology in the broadband networks. A control scheme using neural networks and masked genetic algorithms is proposed and applied to bandwidth allocation of the virtual paths in the broadband networks. The proposed scheme is capable of estimating utilization of the virtual path’s network, and adaptively selecting optimal bandwidths of the virtual paths using masked genetic algorithms according to the arrival characteristics of the multirate’s traffic and network environment. As the optimization problem is constrained, the traditional genetic algorithms are no longer applicable to this problem. The dissertation proposes the masked genetic algorithms to solve the optimization problem. Simulation results demonstrate the effectiveness of this kind of dynamic allocation.· Presenting a new admission control algorithm based on measurements and neural network estimation. Because of the diversity of traffic characteristics and the QoS requirement, it is very difficult to estimate QoS of statistical multiplexing traffic by preciseanalysis model. The method we proposed in this dissertation is based on arriving measurements of aggregated traffics using sliding window and estimation of the neural networks for cell loss rate, it can implement quick QoS estimation and call admission control using characteristic parameters inside the traffic. This method can fully take advantage of statistical multiplexing characteristics an更多还原