Network-monitoring Method based on Self-learning and Multi-dimensional Analysis

摘要

A novel network-monitoring system for detecting abnormal network conditions (such as hidden network congestion) is proposed. The proposed monitoring system is based on self-learning and multi-dimensional analysis. It analyzes multiple parameters such as consumed bandwidth, packet size, and arrival interval of network packets simultaneously. By executing high-quality network monitoring it thereby achieves multi-dimensional analysis by use of Mahalanobis distance. A prototype monitoring system was constructed and evaluated. The evaluation results indicate that the monitoring system can accurately detect a hidden change of network-traffic conditions and reduce the number of unnecessary alerts for monitoring excess bandwidth according to a set threshold.