A Web Client Perspective on IP Geolocation Accuracy

Geolocation of Internet addresses is widely used by content providers to tailor services and content to users and to restrict access to content. IP geolocation in practice typically relies on databases that include geographic information about individual addresses or address prefixes. Prior studies have assessed the accuracy of these databases by using a set of addresses with known or estimated locations and comparing them with database-reported locations.In this paper we investigate IP geolocation accuracy from the standpoint of web clients, by exploiting geolocation information embedded in non-standard HTTP response headers and in unencrypted HTTP cookies. We identify a set of 10,476 websites and content providers that include non-standard HTTP headers and unencrypted cookies with geographic information. We launch HTTP requests to each of these sites from 113 client vantage points with known locations distributed across 6 continents and 60 countries and extract available geographic information from the responses using a battery of hand-crafted regular expressions. We find that the country of the client is included in more than 90% of all responses. Moreover, we observe that about 75% of all responses only include the country name or code and that the remaining responses include some combination of geographic information, such as continent, country, city, postcode, region, and coordinates. We observe that accuracy is greatest for the coarsest geographic scope (continent) and least accurate for finer scopes (e.g., coordinate), but that accuracy varies widely across vantage points regardless of the continent or country from which the request is launched.