A hierarchical Network Coordinate system

Network distance prediction based on Network Coordinate (NC) system provides an efficient mechanism to obtain network distance (latency) information with limited times of measurements. However, the prediction accuracy of single space NC in large network is far from satisfying application requirements, it suffers from higher relative error in short distance estimations especially. In order to improve the short distance prediction, we learn the hierarchical structure feature of the internet and propose a hierarchical predication mechanism based on clustering technology (Clustered-NC). In Clustered-NC, the whole network is clustered into N-level hierarchies, each host is assigned with multiple coordinates corresponding to the N-level hierarchies so that different sets of coordinates satisfy different ranges of distance prediction. Finally, we evaluate the performance of Clustered-NC, and compare it with Vivaldi which is not a hierarchical network coordinate algorithm, results show that Clustered-NC can achieve better prediction accuracy. Many network applications such as nearest server selection will benefit from more accurate prediction using this hierarchical predication mechanism.