Unveiling Internet Censorship: Analysing the Impact of Nation States' Content Control Efforts on Internet Architecture and Routing Patterns

TL;DR

This paper investigates how nation-state censorship shapes Internet topology by integrating diverse data sources to build a comprehensive topology map of the global Internet. It fuses BGP routing data, traceroute measurements, registry metadata, and geolocation to produce a time-sliced AS graph with 82,593 ASes and 176,422 edges for 2023-06-01, achieving 7.1% greater path visibility than prior state-of-the-art baselines. A novel 'funnelling' metric is introduced to quantify how censorship correlates with routing constraints, and the authors illustrate this trend via case studies of the United Kingdom, Iran, India, and Russia, alongside analysis of geopolitical fragmentation over 15 years. The work advances observability, enables cross-country comparisons, and provides a practical framework and dataset for policymakers and researchers studying Internet resilience under geopolitical pressures.

Abstract

Heightened interest from nation states to perform content censorship make it evermore critical to identify the impact of censorship efforts on the Internet. We undertake a study of Internet architecture, capturing the state of Internet topology with greater completeness than existing state-of-the-art. We describe our methodology for this, including the tooling we create to collect and process data from a wide range of sources. We analyse this data to find key patterns in nation states with higher censorship, discovering a funnelling effect wherein higher Internet censorship effort is reflected in a constraining effect on a state's Internet routing architecture. However, there are a small number of nation states that do not follow this trend, for which we provide an analysis and explanation, demonstrating a relationship between geographical factors in addition to geopolitics. In summary, our work provides a deeper understanding of how these censorship measures impact the overall functioning and dynamics of the Internet.

Figures (18)

• Figure 1: System Architecture. The figure shows the ingested data from the RIPE RIS noauthor_routing_2015, RouteViews noauthor_university_1999 and PCH noauthor_daily_2010 route collectors, the RIPE Atlas ripe_ncc_ripe_2015 probes, RIPEstat noauthor_ripestat_2022 and GeoLite2 noauthor_geolite2_2023. This is followed by our data processing pipeline which generates an Internet topology model underpinned by BGP inferencing, which is converted into a labelled graph object for a specified time period.
• Figure 2: Inferring the location of an AS by propagating geolocation data. We propagate the location of IP addresses within the advertised prefixes of an AS to infer the physical location(s) of that AS. We do this both for the observed peer_ip in the global routing table, as well as separately for its advertised prefix ranges.
• Figure 3: Representing Internet topology in graph format. Each AS for which we have collected data is represented as a node, alongside attributes providing metadata (here, showing the country of registration and registered owner) and each adjacency is represented as an edge.
• Figure 4: We observe at least one relationship for the majority of registered ASes (showing 10 highest countries by AS registration).
• Figure 5: Countries with the highest proportion of AS degrees (top 20). This shows the sum of all AS degrees registered within an ISO-recognised country, divided by the sum of all AS degrees.