Internet Host Geolocation Based On Probabilistic Latency Models

The robust and scalable growth of the Internet has allowed value added services that provide enhanced user experience. Offering information based on geographic location to Internet users is one of the newest and notable advancements. Finding the geographical location of the user on the Internet, commonly referred to as geolocation, is one of the challenging problems currently addressed by the research community. Geolocation of Internet hosts is not straightforward, mainly due to the underlying behavior of the IP network. In the conventional telephone network each telephone number has a physical address associated with it. On the contrary, an IP address does not have any notion of location or any physical location associated with it. Dynamic assignment of IP addresses makes the task of geolocation even harder. This thesis focuses on scalable and accurate approaches to finding the physical location of Internet users. We consider the scenario of a user connected to the Internet through a machine, which we refer to as a host. Of the two commonly used approaches, repository-based and measurement-based, we primarily focus on geolocating Internet hosts using the measurement-based approach. Existing measurement-based geolocation approaches are based on simple bounds on latency measurements and do not take into consideration the uncertainty associated with latency measurements effectively. We focus on improving host geolocation based on probabilistic models for Internet latency. In this thesis, we address the following issues: • We provide a systematic study to understand the relationship between Internet latency and distance. The observations from this study motivate us to propose a probabilistic model for the latency to distance relationship that captures the high