Research On Key Technologies In The New Generation Network

The explosive growth of Internet makes it facing a lot of challenges. The current Internet architecture can not solve the problem of routing scalability, and does not native support the needs of mobility and security. Besides, today's Internet can not provide the function of perception and measurement, and the pattern of user behavior is also hard to describe. All these drawbacks of the current Internet that root in its original design principles seriously hinder its future development. For this reason, the theory of the new generation Internet is regarded as one of the most important and pressing research content at home and abroad, including many sub-domains such as the network architecture, network virtualization, measurement and prediction and so on.In order to solve the puzzles facing the current Internet, this dissertation focuses on the research of the new generation of Internet architecture, virtual network mapping algorithm and the prediction algorithm of network traffic. The major works are outlined as follows:1. A new hierarchical network with diversified customer edge access (HNDCE) is proposed, which takes advantages of the hierarchically switched networks and the schemes implement identification and location splitting. In the core network, hierarchical tree topology combined with the Viliant Load-balancing (VLB) structure makes routing mechanism is discarded while the reliability and scalability of the tree topology can be compensated. A flow-level load allocation scheme based on the structure of core network is also proposed. In the customer edge network, a hierarchical mapping system is introduced to achieve identification/location splitting, and support the mobility and security effectively. The proposed architecture can solve the problem of routing scalability, mobility and security which the current Internet faces, and facilitates the network measurement and user behavior analysis.2. A novel virtual network mapping algorithm is proposed based on the structure of the hierarchical network that we presented. The new algorithm divides the original virtual network request into the core network and each customer edge network. In the core network, the transmission path is definite and the network recourses is easy to count since the tree topology is adopted, so the virtual network mapping can be simplified as resources lookup between each edge switching node pair. Then the decomposed sub-requests with reduced constrains can be embedded simultaneously in each customer edge network. Therefore, the computation cost can be reduced. Theoretical analysis and simulation results show that the new algorithm can reduce the time consumed in the mapping process and improve the request acceptance rate.3. Traditional traffic prediction model based on wavelet decomposition can not describe "brusty" characteristic of network traffic and the use of conventional time series prediction method usually based on some assumptions, which may not fit the practical situation. In order to solve these problems, we proposed a novel methodology which relies on wavelet packet transform and fuzzy neural network. We use wavelet packet transform as extension of wavelet transform for long range dependence (LRD) decorrelation and further partition of the high-frequency part of the network traffic, then use fuzzy neural network for wavelet coefficients prediction at each time scale. This model can describe "brusty" effectively and has good adaptation, which can also support on-line and off-line applications. Simulation results show that the proposed prediction method has good stability and prediction performance.