On the Manifold Structure of Internet Traffic Matrix
因特网流量矩阵的流形结构

Currently, traffic matrices have been applied to anomaly detection, traffic forecasting and traffic engineering widely, but existing researches only find the linear structure of traffic matrix. In order to search the nonlinear structure of traffic matrix, a traffic matrix model is constructed and traffic matrix datasets are collected from real Internet backbone Abilene. Using classical manifold learning algorithms, based on measurement data from Abilene find that these traffic matrix datasets with high dimensionality (81 or 121 dimensions) have a intrinsic dimensionality of 5 and have all kinds of manifold structures in low-dimension embedding space, influenced by sampling density and noise data.