Hello,
In collaboration with IODA, we published a *new research report* which analyzes *Mozilla telemetry* to investigate *internet shutdowns in Myanmar, Uganda, Belarus, and Iran *over the last year.
Read our research report here: https://ooni.org/post/2021-investigating-internet-shutdowns-mozilla-telemetr...
The folks from Mozilla also published a blog post about our report here: https://blog.mozilla.org/data/2021/11/08/detecting-internet-outages-with-moz...
Below we share information about the research and some key findings.
*# About the research*
Given how widespread Firefox usage is worldwide, could Mozilla telemetry be a valuable resource for the Internet freedom community to investigate Internet shutdowns?
To explore this question, we analyzed an aggregated dataset of network activity based on Firefox web browser usage worldwide (access to which was provided to us by Mozilla).
To evaluate how useful Mozilla telemetry is for researching Internet shutdowns, we:
* Selected case studies (known shutdown events from January 2020 onwards) * Analyzed Mozilla telemetry * Compared Mozilla telemetry with other public datasets
We checked whether Mozilla telemetry provides signals for:
* Myanmar -> Internet outages following February 2021 military coup * Uganda -> Nationwide internet outage amid 2021 elections * Belarus -> Internet outages amid 2020 elections * Iran -> Internet outages following 2020 elections
*# Summary of key findings*
We provide detailed analysis and findings for Myanmar, Uganda, Belarus, and Iran in our report: https://ooni.org/post/2021-investigating-internet-shutdowns-mozilla-telemetr...
Overall, based on our analysis for these 4 countries, we found:
1) *Mozilla telemetry provides strong signals for high impact internet shutdowns* -> When access to the internet was shut down entirely (such as in Uganda and Myanmar earlier this year), we observed a complete absence of Mozilla telemetry. This corroborates the absence of metrics that we also observe in other public datasets (such as IODA, Google traffic data, etc.) that are used to investigate internet shutdowns.
2) *Mozilla telemetry can also be used to infer lower impact internet shutdowns* -> When an internet shutdown is not "total" (when an internet shutdown doesn't impact all networks and/or regions in a country), it can still be possible to potentially infer that there is some internet disruption from the high percentage of connection timeouts and unreachable connections that are visible through Mozilla telemetry. We observed this, for example, for "lower impact" internet shutdowns in Belarus and Myanmar. Of course, it's important to cross-reference this data with other public datasets (such as IODA and Google traffic data).
3) *The geographical granularity of Mozilla telemetry provides new research opportunities* -> In Belarus, Mozilla telemetry showed a spike in connection timeouts and unreachable connections from multiple locations, suggesting that there may have been some reshaping of the network topology during the August 2020 Internet outage.
4) *Mozilla telemetry does not seem to provide signals for short-lived internet shutdowns (lasting less than 1 hour)* -> In Iran, Mozilla telemetry did not provide signals of the two short-lived internet outages that occurred on 3rd and 11th March 2020 (both of which lasted for less than an hour, and which were visible through IODA data). This is likely due to the fact that Mozilla telemetry is aggregated in hourly buckets, therefore missing short-lived internet outages.
Overall, our analysis demonstrates that *Mozilla telemetry is a valuable resource for investigating Internet shutdowns worldwide*, providing novel insights through geographical data granularity.
We encourage Mozilla to *publish aggregated Mozilla telemetry as open data* to support research & advocacy efforts investigating Internet shutdowns worldwide.
We also share further recommendations in our report: https://ooni.org/post/2021-investigating-internet-shutdowns-mozilla-telemetr...
We thank Mozilla for providing us access to Mozilla telemetry for this research, and for considering our recommendations.
Best,
OONI team.