Research On Technologies Of Traffic Assignment And Prediction For The CABO Network Architechture

New Network Architectures, network protocols and network services are continually proposed and designed to improve the Internet’s performance, reliability, controllability and scalability. CABO("Concurrent Architectures are Better than One"), a newly proposed network architechture, raised many challenging research topics for researchers. And among the topics, the research on traffic assignment and prediction problem in virtual network is of great significance for the basic theory and the supporting technologies of CABO.In this thesis, the features of CABO architecture and the basic principles, technolgies and research stuation of traffic assignment and traffic prediction are introduced. The existing traffic assignment and traffic prediction technologies have several disadvantages.(a)the congestion control technology is a partly solution after problem happens, with no treatment for the essential cause of traffic congestion;(b)the MPLS(Multi-Protocol Lable Switch) traffic engineering is multi-protocol based, so that several born problems limit its application;(c)the accuracy of existing traffic prediction algorithm remains to improve.For the virtual network with bandwidth sensitive service, a traffic splitting and assigning algorithm is proposed. With the designated routing algorithm and multi-routing technology be ing feasibly implemented and deployed, Then the traffic assignment problem can be solved in polynomial time. By improving its constraint set, the traffic can be better assigned. For the virtual network with multiple services, by analyzing the characteristic of hybrid traffic and making prediction of the traffic, with the result of traffic prediction, the input set of traffic splitting and assigning algorithm can be optimized, and be used to traffic assignment. For the periodicity and self-similarity of network traffic, a prediction algorithm based on wavelet transform and combination model is introduced.For the traffic series under different time scale, self-similarity is analyzed and different prediction model is selected for predicting, which achieves a higher accuracy. The innovation points are as follows:Firstly, for the virtual network with bandwidth sensitive service, a traffic splitting and assigning algorithm is proposed. The traffic assignment in single-path network, as finding an optimal mapping from a link to a single path reduces to the Unsplittable Flow Problem, can not accept as many requests as possible. In a virtual network, the routing protocol can be customized. So the designated routing algorithm and multi-routing technology can be feasibly implemented and deployed. Route computing with bandwidth allocation can be considered as a resouce mapping problem. Under flexible splitting over multiple paths by designated routing, the problem is reduced to the MFP. It can be solved in polynomial time. By improving its constraint set, the traffic can be better assigned. The assignment algorithm makes better solution, less computing time and better network performance (bandwidth utility, time delay and package lossrate).Secondly, for the virtual network with hybrid services, a prediction based traffic splitting and assigning algorithm is proposed. It’s hard to get the real-time status of links in the hybrid services network, so that the traffic splitting and assigning based traffic assignment algorithm is not feasible. But with the result of traffic prediction, the links status in the input set can be predicted and reranked. The chosen link will be mostly available, and the traffic splitting and assigning algorithm will be more practical. Simulation results show that the algorithm achieves a higher bandwidth utilization ratio and less package lossrate compared with the non-traffic assighment network.Lastly, for the multi-scale characteristic of network traffic, a prediction algorithm based on wavelet transform and combination model is introduced in this paper. The complex correlation structure of the network history traffic is exploited with wavelet method.For the traffic series under different time scale, self-similarity is analyzed and different prediction model is selected for predicting with SARIMA (Season Autoregressive Integrated Moving Average) models. The result series is reconstructed with wavelet method. Simulation results show that the combination method can achieve higher prediction accuracy lower computation complexity rather than any single prediction model.