Research on Optimizing Embedding Space Dimension in Network Coordinate System

Network coordinate system provides a method to predict distances between network nodes by geometric space embedding, which can help distributed applications such as P2P and Grid select the best node to transfer data. However, how to select the suitable embedding space dimension is still an open problem. In this paper, firstly we give an experimental study on the relationship between distance prediction accuracy and embedding space dimension, and then discuss the cluster feature of Internet distance, which has an important influence on embedding space dimension selection. Then based on such feature, we propose an embedding space dimension optimization selection algorithm. Analysis result shows that this algorithm can select the best embedding space dimension and landmarks, and provide a mechanism for network coordinate system to implement accurate distance prediction with low system cost.