Impact of Network Topology on Distance Prediction Accuracy

Internet applications such as P2P, grid and CDN can benefit much from the ability to get network distance information before data transferring. By embedding Internet distances into a geometric space, virtual coordinate based network distance prediction mechanism provides a method to predict distances between arbitrary network nodes. Firstly the fact that some special network topologies cannot be embedded into a Euclidean space without embedding error is proved. And then a simulation study on the prediction error of different network topology is given, namely triangular inequality satisfaction (TIS) mesh topology and triangular inequality violation (TIV) mesh topology. Simulation results show that network topology has great influence on distance prediction accuracy. When doing space embedding, TIV topology has great prediction error, but meanwhile prediction accuracy of TIS mesh topology is extraordinarily high, which means TIS mesh topology is more suitable to be embedded into a Euclidean space. This conclusion reveals the fact why most prediction error is produced in edge network, and prediction error in core network is small.