Research On Algorithms For Multi-Domain Virtual Network Embedding With Limited Information Disclosure

As the number of users and types of network services increase, functional limitation, capacity constraints and other problems of the conventional network architectures have become increasingly significant. In this context, network virtualization technology becomes an effective way to solve this kind of problem. Network virtualization is a technology that abstracts, reassigns and isolates the substrate networks in order to enable the sharing among different users. And it supports the diversity of the network services. There are so many ways to embed a virtual network into the substrate networks in network virtualization. And the more virtual networks there are on the substrate networks, the more profits the infrastructure providers can get. So how the substrate networks dynamically host the virtual networks to meet the needs of performance indicators is very important. This kind of problem is called virtual network embedding problem, which is one of the key issues of network virtualization.Existing researches of virtual network embedding have focused on the single-domain virtual network embedding. However, with the expansion of the network scale and the diversification of network services’requirements, virtual network requests may need to be embedded into multiple substrate networks, which belong to different infrastructure providers. Embedding makers know the global topology information and resource information of the substrate networks in the existing single-domain virtual network embedding algorithms. But different infrastructure providers have their own network information protection policies of the multi-domain substrate networks. Therefore, the existing single-domain virtual network embedding algorithms can’t be applied to the case of multi-domain virtual network embedding. Based on the multi-domain network scenario, we propose two multi-domain virtual network embedding algorithms to minimize the cost of virtual network embedding and balance the network traffic.The main works and contributions of this dissertation are as follows. (1) A multi-domain virtual network embedding algorithm based on limited substrate network information is proposed. It ensures the confidentiality of the substrate network information and reduces the cost of multi-domain virtual network embedding.The algorithm aims at minimizing the cost of multi-domain virtual network embedding. Local controllers select candidate substrate nodes based on the estimated cost of unit virtual network embedding. After that, the global controller segments the virtual network requests based on the candidate substrate nodes. The multi-domain virtual network embedding algorithm solves the problem of the limitation of substrate networks information. The simulation results show that the algorithm can reduce the cost of multi-domain virtual network embedding, and has good stability and scalability.(2) A multi-domain virtual network embedding algorithm based on traffic balance is proposed. It improves the acceptance rate of virtual network requests under the congestion of substrate networks.Existing researches of multi-domain virtual network embedding have focused on how to minimize the cost of multi-domain virtual network embedding without considering traffic balance of the substrate networks after the completion of the virtual network embedding. The global controller considers minimizing the cost of multi-domain virtual network embedding and traffic balance of the substrate networks when segmenting virtual network requests in the algorithm. The algorithm can realize traffic balance of the substrate networks as the embedding cost is within the acceptable range. The simulation results show that the algorithm can significantly enhance the acceptance rate of the virtual network requests when the cost of the virtual network embedding has not significantly increased in the case of network congestion.