Structured nonnegative matrix factorization for traffic flow estimation of large cloud networks

Network traffic matrix estimation is an ill-posed linear inverse problem: it requires to estimate the unobservable origin destination flows, X, given observable link Y, and a binary routing matrix, A, which are such that Y=AX. This challenging but vital problem as accurate of OD flows required for several network management tasks. In this paper, we propose novel model maps high-dimension low-dimension latent with following three constraints: (1) nonnegativity constraint on estimated (2) autoregression enables proposed effectively capture temporal patterns (3) orthogonality ensures mapping between low-dimensional corresponding be distance preserving. The parameters training algorithm based Nesterov accelerated gradient generally shows fast convergence. We validate flow two real backbone IP datasets, namely Internet2 GÉANT. Empirical results show outperforms state-of-the-art models not only in terms tracking individual also standard performance metrics. found highly scalable compared existing approaches.