A Multi-dimensional Coordinate Factorization Algorithm for Network Distance Prediction

Large-scale Internet applications can benefit from an ability to predict round-trip times to other hosts without having to contact them first. In various predictable model, network coordinates system is an efficient mechanism for internet distance prediction with limited measurements. In this paper, we identify a coordinate matrix which consists of measured value between benchmark node and other common node, so convert computation between nodes into question of factorizing coordinate matrix. We present an algorithm of matrix factorization which factorize coordinate a matrix into non-negative matrix U,V. Through the factorization of the matrix greatly reduces the dimension from the calculation, make for fast convergence of the prediction algorithm.