Characteristics of Internet Latency and Their Impact on Distance Prediction Accuracy

Network distance prediction based on network coordinate (NC) system provides an efficient mechanism to obtain network distance (latency) information with limited times of measurements. In this paper, the mathematic principle of metric space embedding in NC system is analyzed, and the basic rule of designing embedding space model is proposed, which is minimizing both embedding transformation complexity and distortion. Then based on different performance metrics, the performance of a typical NC system is evaluated, and analysis results show that it is far from satisfying application requirements. We ascribe the main causes of such results are Internet hierarchical structure and network distances quasi-metric space feature, the latter of which is then studied in detail. Finally the practicability of increasing network distance prediction accuracy using hierarchical model and space repair technique is discussed.