Revisiting Constraint Based Geo Location: Improving Accuracy through Removal of Outliers

IP based GeoLocation has become increasingly important for network administrators for a number of reasons. The first and foremost is to determine cyber attackers' geographical location to prevent, contain and thwart further attacks. A secondary reason may be to target potential advertisements to users based on their geographical location. IP addresses are indicators of the users location at a course level such as country or city. For further refined estimate of user location within a city, for example, network measurements as simple as simple round trip time of pings to the user can give a more precise estimate of users location. Recently, several researchers have proposed constraint based optimization approaches to estimate a users location using a set of landmark hosts using multi-lateration. In this work, we show that the optimization of such schemes are sometimes plagued by outliers which causes location estimated to be deteriorated greatly. We provide anecdotal evidence for this and proposals to alleviate these problems for detection, masking or removal of these outlier measurements and show that location error improvement of several hundreds or thousands of km are possible.