Research On Traffic Matrix Estimation With Noise

With the rapid development of communication technology and information technology in recent decades, the scale of the network is getting larger, the network structure is becoming more and more complex, more and more users access to the internet. The internet service providers in order to make the construction of network better, design the network, maintenance the network stable, to provide higher quality network services, also need to master the various parameters of the network, traffic matrix which described the whole network in the end-to-end flow distribution, is an important parameter of network analysis. Due to the complexity of network structure and the large of its size, the entire network traffic matrix is obtained through direct measurement is infeasible or cost much. Again because of the link traffic and routing instructions matrix can be obtained with low cost, so how to use link traffic and routing instructions matrix to estimation traffic matrix received increasing attention from home and abroad scientific researchers, it become a hot research problem in recent years.Based on traffic matrix estimation is a reverse pathological problem solving, often mixed with noise in link traffic, traffic matrix estimation algorithm for noise suppression effect is studied. Concrete results are as follows:1. This thesis in view of the existing traffic matrix estimation algorithm always based on prior distribution, and no link traffic contained Gaussian white noise suppression of design to the traffic matrix estimation. The traffic estimation algorithm based on iterative weighted search is proposed in this thesis. Traffic matrix estimation is the solving of a pathological problem, because it is underdetermined. So there are numerous solutions satisfy the linear constraint conditions, so we need to hypothesis the traffic between OD pair. However the method of assuming the traffic between OD pairs satisfies a specific distribution will ignore the time correlation and spatial correlation within traffic matrix data, at the same time distribution assumption is often have a bias which is not accurate. LENS link flow decomposition algorithm is presented in this paper; on the basis of it we add the time constraints, and the space constraints, to optimize the solving process in order to get a better estimation of traffic matrix. We put forward the traffic matrix estimation algorithm based on LENS matrix decomposition. The algorithm in the case of no noise is able to accurately estimate the big traffic OD flow, on the whole traffic matrix estimation precision is higher than that of the mainstream algorithm, at the same time, the algorithm of link traffic Gaussian white noise have very good inhibitory.2. Based on simple gravity model to estimate traffic matrix estimation the result does not comply with the traffic matrix estimation model and the estimation methods of existing issue such as difference of mixed noise suppression capability. The traffic estimation algorithm based on iterative weighted search is proposed in this thesis. Gravity model is one of the most efficient traffic matrix estimation, because it does not use the linear constraint condition of traffic matrix estimation, and it is based on the assumption that the traffic between the OD pairs is independent, which make the estimate result relatively coarse. But the estimates result of gravity model turned out to be a very good estimate of the initial value. So in this paper, on the basis of gravity model we add the linear constraints of the traffic matrix estimation to traffic matrix estimation, through the weighted least-square solutions to solve the traffic matrix. Weighted efficient design used the mixed distribution function, which have a very good inhibitory of the link flow with the Gaussian white noise and impulse noise. This article is not directly to minimize the objective function, but through the residual in the vicinity of the initial value iterative search satisfy the constraint conditions of the solution. Through the experimental simulation show that the proposed iterative weighted search traffic matrix estimation method can well get good result and restrain Gaussian white noise and impulse noise in link.