数据中心网络架构及虚拟网络映射研究

发布时间：2018-12-12 15:29

【摘要】：传统的数据中心网络多采用树形拓扑构建，但网络过载问题严重，同时采用高性能专用设备使得设备成本激增。Fat-tree网络缓解了传统树形结构的过载问题，降低设备成本开销，但仍无法克服树形网络扩展能力受限的固有问题；Fat-tree网络下行链路唯一，不利于下行路径容错。另外，数据中心网络具有多租户特性，可以支撑多个虚拟网络，但现有资源分配策略难以保证服务质量，资源利用率普遍低下。因此，本文研究的重点之一即是如何设计一种继承Fat-tree拓扑优点且克服Fat-tree拓扑缺陷的网络架构，以适用于构建数据中心网络；同时，设计一种能够优化网络资源配置，，保证用户服务质量的虚拟网络映射算法也是本文的研究重点。本文首先对数据中心网络发展现状进行了总结，指出数据中心网络的特性需求，并详细介绍了典型的拓扑架构。接着对现有虚拟网络映射算法的优缺点进行总结。然后设计了一种基于模块化的数据中心网络拓扑架构Torus of Fat-tree(ToF)，旨在解决Fat-tree网络扩展能力受限于交换机端口数量及下行链路缺乏动态选择性的问题，同时给出三种网络扩建方式以支持网络规模的灵活扩建。另外，为提升网络性能，设计了一种分布式路由算法。最后基于OPNET完成仿真性能验证，结果表明，ToF具有良好的网络吞吐、时延性能，同时能够很好地支持局部流量通信模式，适应于构建数据中心。在此基础上设计了一种基于用户优先级的虚拟网络映射算法，用以保证用户服务质量，并给出了模型求解算法，最后构建验证系统，对映射模型的性能进行评估，通过系统验证，虚拟网络映射模型能够很好地满足不同用户的服务需求，实现资源的优化配置，提升资源利用率。
[Abstract]:The traditional data center network is constructed by tree topology, but the network overload problem is serious, and the equipment cost is greatly increased by using high performance special equipment. Fat-tree network alleviates the overload problem of traditional tree structure. Reduce the cost of equipment, but still can not overcome the tree network expansion capacity constraints inherent problems; The downlink of Fat-tree network is unique, which is unfavorable to the fault tolerance of downlink path. In addition, the data center network has the characteristic of multi-tenant, which can support many virtual networks, but the existing resource allocation strategy is difficult to guarantee the quality of service, and the utilization rate of resources is generally low. Therefore, one of the key points of this paper is how to design a network architecture that inherits the advantages of Fat-tree topology and overcomes the defects of Fat-tree topology, which is suitable for constructing data center network. At the same time, designing a virtual network mapping algorithm which can optimize the allocation of network resources and ensure the quality of service of users is also the focus of this paper. Firstly, this paper summarizes the development of data center network, points out the characteristic requirements of data center network, and introduces the typical topology architecture in detail. Then the advantages and disadvantages of the existing virtual network mapping algorithms are summarized. Then a modularized data center network topology architecture Torus of Fat-tree (ToF), is designed to solve the problem that the Fat-tree network expansion ability is limited by the number of switch ports and the downlink lack of dynamic selectivity. At the same time, three kinds of network expansion methods are given to support the flexible expansion of network scale. In addition, in order to improve network performance, a distributed routing algorithm is designed. Finally, the simulation performance based on OPNET is verified. The results show that ToF has good network throughput and delay performance, and it can support the local traffic communication mode well, which is suitable for building data center. On this basis, a virtual network mapping algorithm based on user priority is designed to ensure the quality of service of users, and a model solving algorithm is given. Finally, a verification system is constructed to evaluate the performance of the mapping model. Through system verification the virtual network mapping model can meet the service needs of different users realize the optimal allocation of resources and improve the utilization of resources.
【学位授予单位】：西安电子科技大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TP308;TP393.01

【参考文献】