IPv6 LANMARKER-PLUS : Enhancing Geolocation Accuracy and Efficiency via Staged Probing and Density-Based Clustering

Abstract

The accurate estimation of the street-level geolocation of IPv6 addresses poses a significant challenge due to the vast address space and the rotation of prefixes driven by privacy concerns. Methods like IPv6Landmarker achieve substantial improvement compared to prior IPv4 techniques; however, these rely on brute-force querying and heuristic-based clustering, which are computationally inefficient and vulnerable, respectively. In this paper, we present IPv6Landmarker-Plus-an enhanced extension that significantly improves IPv6 landmark mining in terms of both accuracy and resource efficiency. Our system utilizes Java with a MongoDB architecture to filter EUI-64 addresses out of active IPv6 hitlists and extract associated WAN MACs for cross reference with the Wigle.net database. Our contributions include two major algorithmic enhancements:  1) Staged Offset Probing (SOP): a probabilistic query model that, for high probability offsets such as -1, +5, and 0, reduces the number of external API calls by more than 80%, significantly reducing latency and resource consumption. 2) DBSCAN-Centric Filtering: this replaces heuristic clustering with Density-Based Spatial Clustering to accurately identify the true coordinate cluster while robustly rejecting noise from adjacent network devices, leading to direct improvements in the mean geolocation error. Real-world validation using RIPE Atlas ground truth probes confirms that the DBSCAN-based selection of landmarks achieves a lower mean error and geolocation success rate compared to the original technique in IPv6Landmarker.