节点文献

基于负载平衡的新一代核心网络技术研究

Research on New Generation Core/Backbone Network Technologies Based on Load Balancing

【作者】 杨帆

【导师】 刘韵洁;

【作者基本信息】 北京邮电大学 , 通信与信息系统, 2010, 博士

【摘要】 随着Internet的广泛应用和高速发展,用户对网络的性能、质量和业务的要求与日俱增,网络资源的消耗和网络拥挤日见显著。如何为用户提供多样化的服务质量保障,如何使网络资源分配更合理、利用率更高效,是网络研究面临的挑战。为此,对新型网络体系结构、网络技术和网络设备改造的研究在世界各国纷纷开展,虚拟化网络、负载平衡网络等新型网络架构、网络技术不断涌现。然而,这些新技术目前还缺乏成形的网络解决方案和相关技术算法,如何将他们具体化,推进他们在网络中的应用,是研究的热点。新型网络技术的研究将涉及到网络中的多个方面,随着网络和业务需求的多样性发展,研究所涉及的问题也非常广泛。本论文分别从支持区分QoS的高性能路由器交换结构的研究、基于负载平衡思想的路由策略的研究和虚拟网络与现实物理网络之间的映射算法的研究等几个方面进行了探讨。具体为:1、目前支持区分QoS的高性能路由器交换结构存在中心算法调度、反馈机制、结构复杂等问题。本文利用PPS结构降速特点和负载平衡思想,并参考虚拟化技术思路,设计一种支持区分QoS的高性能路由器交换结构:Weighted-Layer-Assignment Parallel Packet Switch (WLA-PPS)。WLA-PPS基于权重为不同QoS类包流分配相区分的交换层,并仅需用轮转,就实现了分布式、区分QoS支持和降速交换。经理论推导和仿真验证:WLA-PPS可以对不同QoS类包流进行单独的测量、管理和控制,并能在拥塞情况下保证高QoS包流的交换稳定性、带宽和时延性,且比基于中心调度和反馈的结构拥有更好的吞吐量。WLA-PPS实现复杂度低、不需要中心控制和反馈、简易动态可调,为新一代高性能路由器交换结构设计和虚拟化框架下的核心路由器交换结构实现提供了思路,研究成果已申请国家发明专利。2、Valiant Load-balancing (VLB)路由存在全网逻辑全互连(Full Mesh)、中心节点(网)、拓扑单一等问题,以至难以推广。为解决这些问题,本文受蜂窝网中小区划分的结构的启发,设计两种基于VLB的路由算法:Transit-Transit (TT)-VLB路由和Transit-Peering (TP)-VLB路由。TT-VLB和TP-VLB将核心/骨干网络中的节点分为多个小区域,采用不同的逻辑连接和路由方式,仅需局部逻辑全互连、无中心节点(网)要求、适用于广泛的拓扑结构。经理论推导和仿真验证:TT-VLB具有接近VLB的吞吐量、抗毁能力和更优的时延性;TP-VLB保持了一定的VLB性能,拥有较少的跳数、较短的路径。为VLB路由的应用拓展提供思路。3、虚拟网映射算法中存在匹配方程求解复杂、计算花销大、缺乏具体路径选择方法等问题。为解决这些问题提出了利用负载平衡和小区分划的思想,设计一种虚拟网映射算法:Valiant Load-balancing-Virtual Network Embedding (VLB-VNE)。VLB-VNE将核心/骨干网络中的节点分为多个小区域,小区域内采用逻辑全互连和负载平衡路由,小区域间利用最短路径路由和贪婪式迭代匹配算法完成虚拟网映射。经理论推导和仿真验证:VLB-VNE构建合理、易被映射的现实网络,并将节点和链路的映射统一模块化,映射匹配简单,计算花销小、路由路径简易明确,并拥有负载平衡路由的优秀时延和吞吐量性能。4、缺乏一种简易的对PPS等高性能路由器交换结构仿真方法制约着高性能路由器交换结构的研究。为此,提出一种基于NS2软件的搭建PPS结构仿真平台的方法。通过修改NS2软件中的构件库和包转发进程来实现对新型交换结构的仿真模拟,构建起新型高性能路由器交换结构仿真验证平台。这种构建新型交换结构仿真平台的方法扩展性好、开发难度小、简单易于实现。

【Abstract】 With the popularization and rapid development of Internet, high-performances and high-quality services are required increasingly, and resource consumption and congestion are growing significantly. How to provide differentiated services and how to achieve rational resource allocation and efficient utilization are the challenges for the network research. At present, researches of new network architecture, new network technology and new network equipment modification are carrying out in countries of the world. Then network virtualization, load-balancing network and other new network architectures, new network technologies continue to emerge. However, these new technologies still lack of practice solutions and technologies algorithm. How to promote their application in the network is a research focus.The resource of new network technologies involves many aspects of network. With the diversity of network development and user requirements, the issues to be solved involve very extensive ranges. This dissertation focuses on supporting differentiated QoS switch fabric, valiant load-balancing routing scheme and virtual network mapping algorithm. The major works are outlined as follows:1. At present, switch fabrics of high-performance routers which support differentiated QoS have the issues of center controller, feedback and complex architecture. To address these issues, we design a new switch fabric supporting differentiated QoS called Weighted-Layer-Assignment Parallel Packet Switch (WLA-PPS). The idea of WLA-PPS refers to PPS structure, load balance and network virtualization. WLA-PPS assigns different switch layers to different QoS types’packet-flows based on weight values, and achieves distributed control, differentiated QoS support and lower speed switch just by round-robins. Theoretical analysis and simulation results prove that: WLA-PPS can provide measurability, manageability and controllability for different QoS flows separately, guarantee stability, bandwidth and delay for high-QoS flows in the case of overload, and own better throughput than the structure with center controller and feedback. For its low complexity, full distributed implementation and simple dynamic adjustability, WLA-PPS provides an idea for the practicability of both new generation high-performance router and network virtualization core router, and related research has applied for state patent.2. Valiant Load-balancing (VLB) routing scheme has the issues of a logical full mesh, hubs or intermediate nodes (networks) and single application of topology. To address these issues, we propose two novel routing schemes called Transit-Transit (TT)-VLB and Transit-Peering (TP)-VLB. Based on VLB and cell allocation of cellular network architecture, TT-VLB and TP-VLB divide the nodes of core/backbone network into several regions. TT-VLB and TP-VLB use different ways to connect logically and route, then they achieve logical local interconnection, non-central nodes (networks) and a wide range of application of topology. Theoretical analysis and simulation results show that:TT-VLB possesses good throughput and failures performance which are close to the ones of VLB, and it even has better delay performance. TP-VLB maintains a certain amount of VLB performance, and has less hops and shorter paths. TT-VLB and TP-VLB provide ideas for the application of VLB.3. At present, virtual network mapping algorithms have the issues of solving mapping equations complicatedly, high computation cost and lack of specific path selection method. To address these issues, we propose a novel virtual network mapping algorithm called Valiant Load-balancing-Virtual Network Embedding (VLB-VNE). VLB-VNE divides the nodes of core/backbone network into several regions. The topological structure of the regions is logical full mesh, and the matching algorithm of VLB-VNE is based on greedy iterative algorithm. Theoretical analysis and simulation results prove that:VLB-VNE achieves a rational and mapping easily practical network, unites nodes and links mapping, reduces computation cost and possesses optimality of valiant load-balancing network in throughput delay and routing complexity.4. The lack of a simple simulation method negatively affects the evolution of switch fabrics in high-performance routers. In this dissertation, we propose a novel and practical way to evaluate the performance of PPS based on NS2. In this way, PPS can be a module extension for NS2 library by deriving from base classes and modifying the procedure of packet forwarding in NS2. By this method, new high-performance router switch fabric simulation platform can be easy to implement and expand.

节点文献中: 

本文链接的文献网络图示:

本文的引文网络