节点文献

光网络交换技术及其资源优化分配问题的研究

The Study on the Optical Switching Technology and It’s Resource Deployment Optimization Problems

【作者】 单广军

【导师】 朱光喜;

【作者基本信息】 华中科技大学 , 信息与通信工程, 2009, 博士

【摘要】 当前,宽带通信发展迅猛,单纯地扩容硬件设备或增加基础设施投入从经济角度考虑不是一个行之有效的万能办法。光网络领域更是如此,因为用于预测半导体器件发展速度的摩尔定律似乎已经无法跟上人们对带宽需求的增长速度。光通信目前有两个主要的研究方向:光子器件和光网络交换技术。光子器件研究方向致力于实现光包交换(OPS),主要侧重于全光波长变换器、光缓存以及光逻辑单元等半导体器件开发和光信号处理。光网络交换技术主要侧重于适用于大容量、高速光交换网络的网络体系结构设计和协议开发。这两个研究方向相辅相成,共同推进着光通信技术不断地进步。本论文主要针对大容量、高速光交换网络开展网络资源优化的相关研究,包括:路由和波长算法(Routing and Wavelength Assignment:RWA)、冲突规避(Contention Avoidance)以及新型光交换网络体系结构等,以期实现光交换网络的流量负载均衡,降低数据丢失率,并提高系统的吞吐量及资源利用率,针对语音、数据、多媒体各种业务不同的QoS要求给用户提供满意的服务质量。本文首先研究了路由和波长分配(RWA)问题,RWA是一个NP-完全问题,之前的工作尝试以启发式算法或图论来解决这个问题。然而,计算复杂度会随着问题的规模呈指数型增长。特别是在动态的流量模型下,RWA问题变得更复杂,同时阻塞率的估计模型也相当的不准确。有鉴于此,我们提出了一种基于关键链路预测的动态自适应权重RWA算法,我们发现网络性能的劣化往往是从网络中的某些“关键链路”首先开始的,而关键链路的形成又和静态因素(如具体的网络拓扑)以及动态因素(如波长耗尽速度、空闲波长数等)相关。我们提案基于回避网络中关键链路的出发点,并综合考虑跳数和空闲波长数等因素做为权重,动态地为连接请求分配最优路由和波长。相对于常规的RWA算法,在具有低计算复杂度的同时可以实现较好的流量负载均衡,更低的阻塞率。接下来我们分析和完善设计了一种基于时隙的光网络交换技术—光时隙交换(OTS)技术。传统的光突发交换网络(OBS)由于光突发包长为变长尺寸,封装时间也不确定因素,因而是一种面向非连接的光网络技术,这种结构使得在网络负载较重的场合,突发包很容易发生竞争冲突而产生数据丢失,其在资源合理预约、QoS保证以及有效降低网络阻塞率等方面存在一些困难。OTS继承OBS的控制机制,但在时间域内将波长资源分割成一系列固定大小的时隙槽,并以此作为基本的光网络交换粒度。文中并给出了光时隙交换和光波长变换在避免数据冲突方面具有等价性的数学分析证明,通过设置合适的时隙大小和帧长度,作为解决光突发数据包冲突竞争的一个方案,光时隙交换可以实现乃至超过光波长变换的作用。OTS技术有能力通过周期性地预约时隙槽的位置,以虚连接的方式提供面向连接的服务。本文还对光时隙交换器的结构进行了设计,有效地降低了其结构实现复杂度。然后我们就OTS技术中的路由、波长和时隙分配(RWTA)问题进行了研究。几种有效率的时隙分配和调度方法,如Train approach、Wagon approach和p-distribution approach被提出。一个新的自适应权重因子概念被引入到路由和波长选择算法中,以根据网络资源使用情况动态地选择路由和波长。仿真结果表明,我们的提案相对于传统的OBS技术,可以提供更好的服务质量保证(QoS),并且在降低网络阻塞率方面的具有显著优势。最后本论文提出了一种适用于光突发交换网络(OBS)的新型冲突预先规避机制。相应于OBS网络中的每个入口边缘节点,我们都会根据流量情况为其分配一个或一个以上的专用波长,突发包将沿着专用波长通道传送至不同的目的节点。通过这种方式,不需配置波长变换器或光纤延迟线(FDLs),OBS网络中核心层结点上的冲突就可以被部分地规避,而发生在入口边缘节点处的冲突则可以通过缓存予以解决。进一步的,我们提出了一种流量负载均衡扩张树算法,它能根据入口边缘节点的度数均衡地将流量分布于各生成子树之间。在同一生成树内,去往不同目的节点的突发包可以使用同一波长通道而不会发生冲突,得以实现对波长资源的有效利用。仿真结果表明,相对于其它算法,在无需配置波长转换器的情况下,我们的方案可以明显改善网络丢包率的性能。

【Abstract】 Activity in the area of broadband networking has been expanding at a very rapid rate, it seems the Moore Laws’s, which is used to forecast the development trend of semiconductor industry, can’t catch the step of people’s need in the network broardband field. Therefore, purely adding the hardware or increasly the investment in the infrastructure isn’t an efficient way. Currently, it has two main research branches in optical communication field: photonic device and optical networking. Photonic technology mainly focus on the proble how to implement the pure optical packet switching (OPS), which involves the all-optical wavelength converter, optical buffer, optical logical switcher and optical singal processing etc. Optical networking technology mainly study the system architecture and protocols for the large-capacity and high-speed optical switch network. Above-mentioned research branches closely tie together, and continuously push the progress of optical communication field. Our work mainly study the resource optimization problems in the large-capacity and high-speed optical switch network, including RWA ((Routing and Wavelength Assignment) algorithm, contention avoidance mechanism, and novel optical switching network architecture, its objects are to balance the traffic load, decrease data loss rate, improve the network throughput and resource utilization, moreover, provide the good service quality for voice, data or multimedia businesses.Firstly, we take the routing and wavelength assignment (RWA) problem into account. Former works try to solve this problem via heuristic algorithm or graph theory. However, the complexity of the optimization problem grows exponentially as the the size of network grows. In particular for dynamic traffic model, the solution for RWA problem is not only more difficult, but also inexact. In view of these circumstances we propose a dynamic weight RWA algorithm which base on the principle of "Key Link" forecast. We consider the network performance deterioration very possibly derive from certain key links of network firstly, and these key links are related to static factor, such as particular network topology, or dynamic factors, such as wavelength exhausted rate as well as the number of free wavelength. One of most important idea in our proposal is that these "Key Links" should be avoided in advance as possible, then comprehensively takes the hops and free wavelengths into account in order to dynamically choose and assign the optimal route and wavelength for arrival connection requests. Comparing to conventional RWA algorithms, our algorithm can achieve better load balance and lower blocking probability with less computing complexity.Next, a novel optical switching technology called optical time-slot switching (OTS) is presented in charpter 3. Conventional Optical Burst Switching (OBS) network has inherited disadvantages in the aspect of resource reservation mechanism, QoS guarantee and burst contention is more likely to occur in the case of heavy load due to both burst size and assembled time are uncertain in it. With similar control plane mechanism in OBS network, OTS technology divides the wavelength resource into fixed size time-slots as the basic switching granularity of optical network. Also, we mathematically prove the time-slot switching has equivalent function to eliminate data conflict relative to wavelength converter. If the time-slot size and frame length can be set appropriately, as one of promising solution to decrease data contention, optical time-slot switching might reach even exceed the function of wavelength converter. Moreover, OTS technology is capable of providing the virtual connention service via periodically booking the certain time-slots in each frame. We also design a novel architecture for optical time-slot interchanger in this charpter, which can effectively decreae the realization complexity of optical time-slot interchanger.In charpter 4, the routing, wavelength and time-slot assignment problem is discussed. Several effective approaches for time-slot assignment and scheduling, such as Train approach、Wagon approach and p-distribution approach, are proposed in this charpter. Also, a novel adaptive weight factor is introduced into the routing and wavelength selection algorithm in order to dynamically choice the routing and wavelength according to the network resource usage status. Simulation results show our scheme can provide good Quality of Service (QoS) comparing to traditional OBS technology, and has great advantage in decreasing the network blocking probability.A novel proactive contention avoidance mechanism used in OBS network is considered in charpter 5. According to traffic status, one or more than one dedicated wavelength is assigned to each ingress node in OBS network, and bursts will transmit along the dedicated wavelength channel to the destination nodes, thus the potential contention at intermediated nodes can be eliminated partly without wavelength converter or optical delay lines (FDLs), the contention at ingress nodes can be solved via electronic buffer. Further, we construct a traffic load balance spanning tree for each ingress node, where input loads are appropriately balanced among sub-trees according to the degree number of ingress node. Within the spanning tree of identical node, bursts never meet contention, and the wavelength resource ultilization rate also maintains ast a high level. Experimental results show our scheme can improve the network performance of burst drop rate significantly.

节点文献中: 

本文链接的文献网络图示:

本文的引文网络