belarus-2020-2021

Belarus 2020--2021

This repo contains data analysis related to the Tor blocking events in Belarus from 2020-2021. In particular, in late February 2021, the censor apparently enumerated the set of obfs4 bridges that were distributed via email and blocked those bridges. Given the set of 1890 bridges that were distributed in February 2021 prior to the 22nd and the subset of 93 email-distributed obfs4 bridges, our goal was to detect based on low connection counts from Belarus that the 93 email-distributed obfs4 bridges were blocked, while avoiding false detections that the other bridges were blocked.

Our main finding is that the 93 email-distributed obfs4 bridges that were blocked were not commonly used in Belarus prior to the censorship event. Very low connection counts (e.g., 0) were common, so we were unable to infer censorship of these bridges based on this signal.

Reproducing our results

Dependencies:

• bash
• curl
• python3
• numpy

To reproduce our results, just run ./run.sh. This will take a long time (about 12.5 hours on my device) and require a few GB of free space. The reason is that it downloads all extra-info records for all bridges from 2020-07 to 2021-04 from the Tor Project's CollecTor service (734 MB compressed, about 20 GB uncompressed), extracts that data, and obtains the needed information (bridge fingerprint, record date, and connection count from Belarus) from each record. We reduce the disk space required by extracting only one archive at a time, then deleting the resulting files after processing. Nevertheless, there are A LOT of these records (about 4 million), and it takes a long time to process them all.

Alternatively, you can run ./run.sh --fast, which starts with a 6.7 MB archive (87 MB uncompressed) containing only the information we need. This is much faster (1-2 minutes on my device) and requires much less disk space (around 250 MB).

Full results will be available in the output file, and the relevant summary will be printed to the console.

Julian dates

Note: The Julian date conversion in get-bridge-data.sh and get-stats.py is 1 day earlier than reported by, e.g., https://aa.usno.navy.mil/data/JulianDate. What matters is that these two files use the same representation of a date, which they do.

License

This project as a whole is licensed under the MIT (Expat) License. It uses data from the Tor Project that is provided under the CC0 Deed.