The contents of this legacy page are no longer maintained nor supported, and are made available only for historical purposes.

Bibliography Details

M.J. Arif, S. Karunasekera, and S. Kulkarni, "GeoWeight: Internet Host Geolocation Based on a Probability Model for Latency Measurements", in Australasian Conference on Computer Science (ASCS), Darlinghurst, Australia, Australia, 2010, pp. 89-98, Australian Computer Society, Inc.

GeoWeight: Internet Host Geolocation Based on a Probability Model for Latency Measurements
Authors: M.J. Arif
S. Karunasekera
S. Kulkarni
Published: Australasian Conference on Computer Science (ASCS), 2010
Entry Date: 2010-10-22
Abstract: Knowing the geographical location of an Internet host is of importance to many of today's Internet services. In this paper we focus on geolocating Internet hosts based purely on latency measurements. Existing latency measurement-based geolocation techniques use the observed latencies from multiple landmarks to the target host to determine maximum bound or both the maximum and minimum bounds of the geographical region where the target host is located. Due to the large variance of Internet latency measurements, the region constrained based on such maximum-minimum bounds tends to be relatively large resulting in large estimation errors. We propose a geolocation algorithm, GeoWeight, which improves the geolocation accuracy by further limiting the possible target region by dividing the constrained region to sub-regions of different weights. The weight assigned to a subregion indicates the probability of the target being in that sub-region; a higher weight indicating a more probable region. By considering latency measurements from multiple landmarks and computing the resultant weights of overlapping regions a better constrained target region can be obtained. This paper presents the GeoWeight algorithm and evaluates its performance using both synthetic and real data by geolocating target hosts in North America. We compare GeoWeight with two popular geolocation techniques, Octant and CBG, by geolocating the same set of targets. The results show that the GeoWeight algorithm outperforms existing techniques.