Exploring Unlabeled Network Monitoring Data in Large-Scale Active Monitoring Systems
Miloš Nastić, Pavle V. Vuletic · 2025
Computer networks produce large amounts of monitoring data daily. Detailed analysis of this data can enable more efficient network operation and management. There are few large-scale active monitoring systems, with tens to thousands of probes scattered throughout the internet which gather various network performance metrics. This paper analyzes performance metrics gathered from two such systems: GÉANT PMP, based on perfSONAR, and RIPE Atlas, and analyzes the reasons for monitored network anomalies without the information about the events in the underlying network. This study confirms that, within the same time frame and across the same autonomous systems, measurements of the same type have nearly identical values from both platforms proving the reliability of their measurements. However, the exploratory data analysis, although able to find a correlation between the measurements in specific time intervals, shows limits in the ability to interpret the reasons for those anomalies.