光网络虚拟化支持网络云业务关键技术研究

发布时间：2019-01-05 08:23

【摘要】：近年来,网络云业务发展迅速,导致互联网流量出现了指数级增长。传统光网络体系架构在可扩展性和灵活性方面的问题凸显,网络“僵化”的问题越来越突出。为有效解决该问题,研究人员提出光网络虚拟化技术。该技术基于抽象、分区与聚合等机制,在共享的底层网络上构建多个具有异构性、共享性和隔离性等特点的虚拟网络(VN),以此满足网络云业务的异构化网络需求。然而,VN必须依靠底层光网络资源才有意义,因此VN到底层光网络资源的映射问题成为了当前的一个研究热点。研究发现,网络云业务对底层光网络资源需求的主要特点是具有动态性。因而如何动态、高效的将VN映射到底层光网络资源上,成为了决定网络云业务服务质量的关键。本文首先对基于SOA的网络云业务模型的虚拟化层进行了模块化设计,并将其作为后续映射算法的验证环境,使本文研究的验证环境更加接近真实的网络云业务环境。而后,基于光网络虚拟化面临的主要挑战,对面向资源需求量增加和单节点失效两种情况下的动态VN映射问题展开研究。在面向资源需求量增加的动态VN映射算法研究中,研究了虚拟节点资源需求量增加和虚拟节点数量增加两种情况,利用混合整数线性规划(MILP)方法建模。在启发式算法设计中,采用最小节点压力优先方法,充分考虑了重配置过程中底层物理节点的负载均衡。仿真结果显示,相比对比算法,该算法能够有效降低重配置成本,同时实现较好的物理节点负载均衡。在面向单节点失效的动态VN映射算法研究中,主要研究单物理节点失效情况下,其上承载虚拟节点的重配置问题,利用整数线性规划(ILP)方法建模。启发式算法设计中,综合考虑虚拟网络拓扑和底层物理节点的连通性问题,并根据连通性计算结果,选择最佳的候选节点进行映射。仿真结果显示,本算法虽相比对比算法消耗较多的映射时间,但提升了失效VN的恢复率以及降低了底层物理节点的负载强度,因此具有可用性。
[Abstract]:In recent years, network cloud business has developed rapidly, resulting in exponential growth of Internet traffic. The problems of scalability and flexibility of traditional optical network architecture are highlighted, and the problem of "rigid" network is becoming more and more prominent. In order to solve this problem effectively, the researchers put forward the virtualization technology of optical network. Based on the mechanisms of abstraction, partition and aggregation, a number of virtual network (VN), with the characteristics of isomerism, sharing and isolation are constructed on the shared underlying network to meet the needs of the isomerization network of network cloud services. However, VN has to rely on the underlying optical network resources to make sense, so the mapping from VN to the underlying optical network resources has become a hot research topic. It is found that the main feature of the network cloud service is the dynamic nature of the underlying optical network resource demand. Therefore, how to dynamically and efficiently map VN to the underlying optical network resources has become the key to determine the quality of service of network cloud services. Firstly, the virtualization layer of network cloud service model based on SOA is modularized and used as the verification environment of the subsequent mapping algorithm, which makes the verification environment of this paper closer to the real network cloud service environment. Then, based on the main challenges faced by optical network virtualization, the dynamic VN mapping problem in the case of increasing resource demand and single node failure is studied. In the research of dynamic VN mapping algorithm for increasing resource demand, the resource demand of virtual nodes and the number of virtual nodes are studied. The mixed integer linear programming (MILP) method is used to model the model. In the design of heuristic algorithm, the minimum node pressure first method is used to fully consider the load balancing of the underlying physical nodes in the process of reconfiguration. Simulation results show that the proposed algorithm can effectively reduce the cost of reconfiguration and achieve better load balancing of physical nodes compared with the contrast algorithm. In the research of dynamic VN mapping algorithm for single node failure, the reconfiguration of virtual nodes on the single physical node is studied, and the integer linear programming (ILP) method is used to model it. In the design of heuristic algorithm, the topology of the virtual network and the connectivity of the underlying physical nodes are considered synthetically, and according to the results of connectivity calculation, the optimal candidate nodes are selected for mapping. The simulation results show that the proposed algorithm consumes more mapping time than the contrast algorithm, but it improves the recovery rate of the VN failure and reduces the load intensity of the underlying physical nodes, so it is available.
【学位授予单位】：南京邮电大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN929.1

【参考文献】