This article by Geoff Huston examines the inherent challenges of IP address geolocation, which are significantly amplified by the rise of global satellite internet services like Starlink. The author details a project by APNIC Labs that attempts to measure national ISP market share by correlating large-scale online ad impression data with geolocation and network information. However, this model relies on the assumption that users and their IP addresses are geographically static within a single country. Starlink's service—used by mobile entities like ships at sea and by customers circumventing local regulations—breaks this assumption, leading to major data anomalies. Using Yemen as a prime example, the analysis shows how Starlink was erroneously calculated to have a 60% market share. The article identifies similar distortions in over 20 countries and explores the fundamental questions these services raise about the purpose and future of geolocation for mobile and global users. As a pragmatic solution, APNIC Labs has opted to reclassify the anomalous Starlink geolocation data to avoid skewing national ISP statistics.
The Challenge of Geolocation in the Era of Starlink
The Core Problem of Geolocation: The Internet was not designed with a geographic address plan. Mapping an IP address to a physical location is a complex process that relies on external databases. These databases are used for cybersecurity, enforcing intellectual property rights, and national statistics.
Measuring ISP Market Share: APNIC Labs developed a method to estimate ISP market share by combining UN population data, ITU internet user statistics, and a massive dataset of ~35 million daily ad impressions. The model assumes that the distribution of ad impressions correlates with an ISP's user base within a country.
Inherent Model Flaws: This measurement technique has an estimated uncertainty of up to 15% due to major assumptions, namely that Google's ad placements are uniform across a country and that each person uses only one ISP.
Starlink Disrupts the Model: The system's primary assumption—that an IP address's use is confined to a single country—is invalidated by satellite services. Starlink provides service to users who are not geographically fixed, such as ships in international waters or users in countries where the service lacks regulatory approval.
Case Study - Yemen Anomaly: The APNIC Labs model produced a highly improbable result for Yemen, assigning Starlink 6.2 million users and a 59% market share. This is likely caused by two factors:
Maritime Traffic: A high volume of shipping traffic in the Red Sea using Starlink is geolocated to Yemen.
Cross-Border Use: Users in nearby countries without official Starlink access (e.g., Saudi Arabia) may be using roaming services registered in Yemen.
A Widespread Issue: The analysis found that this is not an isolated problem. In 21 countries (including Tuvalu, Sudan, and Myanmar), Starlink's calculated market share is questionably high (over 10%), often due to maritime traffic or users circumventing local regulations by using roaming plans.
Fundamental Questions Raised: Starlink's service forces a re-evaluation of what geolocation should mean. For a user on a ship, should their location be defined by the ground station, the ship's flag state, or the satellite operator's country? The article questions whether static geolocation for mobile IP addresses is still a viable concept.
Key Takeaway & Solution: APNIC Labs concluded that Starlink's geolocation data is skewing national ISP measurements. To correct this, they have decided to override the data for the 20 most affected countries, assigning an "unclassified" designation to these ad impressions to prevent them from distorting the results.