Studies On Key Technologies Of Resource Management In Network Virtualization Environment

After40years of development, the Internet has been deeply rooted in the people’s production and living in various fields. As fully demonstrating its strong vitality and broad prospects for development, the Internet at the same time greatly promoted the economic development, social progress, scientific and technological innovation. With the soaring of user number and the explosive growth of information, as well as the user’s demand on the network shifting, these introduce variety of new network technologies and network services constantly emerging and the scale of network rapidly expanding. The traditional model and architecture of Internet is difficult to adapt to the needs of the future network development, for it is facing a series of challenges in its scalability, mobility, quality of service, security, controllable and manageable, and green energy-saving aspects.Based on the VN management agent and virtual resource agent in virtualization management layer, this paper first studied the resource management architecture in network virtualization environment; And then researched the virtualized resource management and allocation problem from two aspects of virtual network mapping in virtual network instantiation stage and dynamic resource allocation among the running virtual networks; Finally, the thesis designed and the realized the mapping algorithm of the virtual experimental network instantiation for Tsinghua University Network Innovation Environment. Specifically, it mainly includes the following several aspects:(1) In view of the new features of network architecture and the change of relationship between resource supply and demand in the network virtualization environment, this thesis proposed a centralized hierarchical tree-like resource management framework based on the virtualization management layer. It roles as a connecting link between virtual networks and substrate networks, manages virtual network’s resource demand and substrate network’s resource supply, and coordinate the balance of relationship between resource supply and demand. This framework gives a resource management model in the future network virtualization environment. After the introduction of the framework’s hierarchical structure and functional module, the thesis discusses the architecture characteristics in terms of resources’centralized management, distributed control’, in the aspect of decision-making’unified programming, separated deployment’. The analysis shows that the framework can establish good communication and coordination mechanism between substrate virtualized resources and virtual networks to guarantee the substrate resource reasonable utilization and the virtual network efficient construction and operation.(2) In view of the in-depth research of virtual network construction as the first thing in network virtualization environment, the thesis proposed a novel virtual network embedding algorithm based on the minimum cost multi-commodity flow (MCF) model. In the modeling process, the thesis transform the virtual network embedding as a complete flow mapping problem in the MCF model by transferring the node mapping cost into the link mapping cost based on the augmented network graph, which constructed by the physical network and virtual network together. This changed the two-phase situation of first node mapping and then link mapping and only link mapping dependency on node mapping result while node mapping not cooperated with link mapping, achieving unified coordination of node mapping and link mapping for global virtual network embedding optimization. In the modeling process, by definition the embedding cost function with the resource unit price inversely proportional to the remaining resources, the embedding algorithm realizes the load balance of substrate network in virtual network construction process, at the same time by supporting path splitting, the algorithm can effectively use the patch bandwidth resources. The simulation results show that the algorithm can improve the successful virtual network access rate, reduce the cost of virtual network embedding.(3) In view of the resource demands of virtual networks will be dynamic changes with the change of services running on the virtual networks, the thesis proposed dynamic resource allocation scheme among virtual networks in the network virtual environment. By analyzing the features of bandwidth allocation among virtual networks, the scheme establish the bandwidth allocation optimization model based on the multi-objective linear programming (MOLP). Then from the macro adjust and control level of substrate network resource, the scheme transformed multi-objective programming in single objective programming according to the utility optimization model and Quality of Service priority in each virtual network by weighted summarizing objective functions of corresponding virtual networks. Using perception technology for real-time monitoring the resource utilization and virtual network operation state, the scheme adjusted the algorithm’s input parameters and weighting coefficients, and made the bandwidth allocation optimization towards the decision maker’s preferring direction. The simulation results show that, the dynamic bandwidth allocation scheme effectively improves the bandwidth resource utilization efficiency and service satisfaction.(4) The thesis design and implementation embedding scheme in virtual network instantiation stage for the virtual network experiment platform (TUNIE). With analysis of the characteristics and the requirement of virtual network construction in virtual experiment platform, the thesis proposed virtual experiment network embedding scheme based on shuffled frog leaping algorithm (SFLA). Through the simulation analysis shows that the convergence of the algorithm, convergence speed. Finally the scheme is realized in JAVA language and integrated it into the platform management system for virtual experimental network construction. In the platform, the embedding scheme is fully competent the platform’s demand of virtual experiment network instantiation and achieved good performance.