Improving IODA's capabilities to monitor Internet connectivity shutdowns

As of August 2021, the IODA project has moved to Georgia Tech with PI Alberto Dainotti.

Sponsored by:
US Department of State, Bureau of Near Eastern Affairs (NEA)

Principal Investigator: Alberto Dainotti

Funding source:  S-NEAAC-21-CA-3006 Period of performance: March 23, 2021 - September 30, 2022. (Funding actually ended as of August 2021)

Project Summary

This proposed project aims to empower the Internet freedom community to more rapidly and adeptly uncover and respond to Internet censorship events around the world. Specifically, this project aims at improving the accuracy, coverage, usability and sustainability of a system to continuously monitor the Internet and provide in near real-time detailed data about Internet connectivity shutdowns—including data at sub-national and per-ISP granularity. We will also engage with the Internet freedom community in order to improve the usefulness and applicability of our data and tools, and specifically for their application to censorship evasion. While targeting all episodes of connectivity disruption, our collaboration and analysis efforts will especially focus on selective disconnections such as those happening in Iran and on understanding how to maximize insight obtainable through our system in support of censorship circumvention efforts.

Project Milestones

Task 1: Improve IODA’s technical capabilities and sustainability
1.1 Improve our S-ARIMA-based outage detection module ongoing
1.2 Leverage CDN data to detect outages in cellular networks ongoing
1.3 Add support for additional data sources in the Explorer interface ongoing
Task 2: Improve IODA’s usability
2.1 Create basic version of IODA UI ongoing
Task 3: Outreach
3.1 Engage in outreach and collaboration with censorship circumvention and Internet freedom groups and release timely reports about shutdown events ongoing

Proposal Tasks

Task 1: Improve IODA’s technical capabilities and sustainability

This task aims at improving the accuracy, coverage, effectiveness and sustainability of IODA. By making our inferences more accurate, we will be able to better inform the Internet freedom community—and in particular censorship circumvention efforts—in a timely manner. Censorship circumvention tools indeed benefit from a deeper understanding of how, where, and when network interference happens. The improvements we plan to make to IODA’s coverage of events are largely targeted towards connectivity disruptions affecting cellular operators. Such operators often employ Carrier Grade NAT and for this and other various technical reasons (e.g., lack of disruption at the routing plane; weak Internet Background Radiation signal) it is hard to observe, detect, characterize, and analyze shutdowns affecting them. Nevertheless, shutdowns of cellular operators are among the main means of government-ordered network interference. Our analysis of new data sources, in collaboration with a large content delivery network (CDN) operator, has been proven a successful strategy: for example, in November 2019 and in early 2020, we identified several outages affecting cellular operators in Iran. In this task we intend to make further progress in applying this type of data. In addition, in this task we propose improvements to IODA’s architecture that will enable the addition and combination of more data sources, as well as safeguard its maintainability thus improving its effectiveness. Such improvements also significantly reduce the effort to operate, maintain and upgrade IODA’s operations and thus have a positive impact on its sustainability.

1.1 Improve our S-ARIMA-based outage detection module

This sub-task aims at improving IODA’s accuracy and support for additional data sources. In the last two years, we have developed—in collaboration with University of Strasbourg, France and the Internet Initiative Japan (IIJ) Research Labs—an automated outage detection method based on Seasonal Auto Regressive Integrated Moving Average (S-ARIMA) models. In 2019, we published a paper at the TMA network measurement conference where we proposed and evaluated this method with actual outage data. In that work we proved the potential of this approach to dramatically increase the accuracy of IODA’s outage detection applied to its Internet Background Radiation ”connectivity liveness signal” (IBR). Subsequently, we successfully deployed a proof of concept to work on IODA IBR country-level signals. We now intend to proceed towards tuning and deploying this approach also for the BGP and Active Probing signals and—for all three signals—not only for countries but also for sub-national geographic areas (e.g., provinces) and individual telecom operators (Autonomous Systems—ASes). In addition, since we have started importing data sources from other organizations and projects (e.g., Google Transparency Report), we will experiment with supporting S-ARIMA-based automated detection also for these data sources.

We will tune and test our proof of concept outage detection module based on S-ARIMA models to work with the broad and diverse set of time-series data we have in IODA. This work will include first extending the existing IBR (network telescope) S-ARIMA monitoring to include data at sub-national granularity and per-network operator. We will then adapt our algorithm implementation and parameters to deploy this approach also for the BGP and Active Probing signals (also at country, sub-national, and per-operator granularity). Finally, since we have recently started importing data from other organizations and projects (e.g., Google Transparency Report data), we will start experimenting with supporting S-ARIMA-based automated detection also for these new data sources. These tasks are well understood, since we already deployed a first proof of concept that works with country-level IBR time series.

1.2 Leverage CDN data to detect outages in cellular networks

IODA currently has limited visibility into the connectivity of cellular ASes, largely because they often use Carrier Grade NAT. However, in several countries, shutdowns of cellular data connectivity are the main means of government-ordered network interference. We have recently started investigating the potential of other data sources to provide visibility into outages in cellular ASes. Specifically, we started collaborating with a large CDN operator to investigate the potential of post-processed data extracted from their logs to provide useful insight into these outages. This collaboration allowed us to observe shutdowns that were not visible in IODA’s three ”connectivity liveness” signals (BGP, Active Probing, IBR), such as shutdowns of AS51684 (Asiacell) in Iraq and AS9498 (Bharti Airtel) in India. These analyses were done by examining (and in some cases importing into IODA) data limited to few operators and only to the time frame of known events. Recently, this collaboration allowed us to investigate in detail and gather significant insight about outages happened in Iran earlier this year (from February 17 to April 17, 2020—a period covering events such as the legislative election in February 21 and the early spread of COVID-19 cases in Iran).

In this task, we intend to move this explorative collaboration to the next phase, in which we plan to import in IODA data from the CDN related to many cellular operators worldwide and apply, tune, and evaluate our S-ARIMA-based outage detection method. We will continue our existing collaboration with Verizon Edgecast and transition from exploration of (and comparison against) their CDN data in few specific cases of Internet shutdowns, to importing longitudinal data for many cellular operators worldwide. We will then continue collaborating with Verizon Edgecast and with the Internet Initiative Japan (IIJ) Research Labs to apply, evaluate, and adapt our S-ARIMA approach to time-series extracted from Verizon’s data. This task is based on our existing collaborations with Verizon and IIJ. This task contributes towards improving IODA’s coverage of events and towards introducing an additional data source.

1.3 Add support for additional data sources in the Explorer interface

In this task, we pair (i) adding support for additional data sources and annotations in the Explorer interface to improve users’ ability to discover, analyze, and document shutdowns, with (ii) design and system improvements to significantly increase the overall project sustainability. The availability of data from distinct sources and of different type is key for a more accurate, deeper, and comprehensive understanding of Internet shutdowns. In addition, the agreement of different data sources increases the credibility and authoritativeness of results, which has important implications, e.g., for advocacy and to inform circumvention efforts. We have recently started importing other data sources related to connectivity shutdowns (e.g., per-country Google Transparency Report time series data). In parallel, we started envisioning and working on radical changes to IODA’s system architecture in order to largely simplify maintenance and reduce the efforts required to introduce new features. For example, we are replacing its time-series database (DBATS, developed by us in 2013) with InfluxDB. However, the current advanced ”Explorer” interface in IODA is tightly coupled with components of our backend (a heavily modified Graphite implementation and DBATS) and would require significant updates to its APIs and javascript frontend (again, custom code that we developed between 2013 and 2016) in order to support additional data sources. We propose to rebuild and extend the Explorer interface by (i) leveraging existing industry-standard open-source projects such as Grafana and InfluxDB, (ii) significantly reducing the custom software modules and code base, (iii) integrating the functionalities that are needed in order to support the core Explorer functionalities with any data source as well as annotations. This sub-task will extend support for additional data sources and improve IODA’s effectiveness and sustainability.

We will rebuild and extend the Explorer interface. We will first migrate data from our DBATS timeseries databases over to InfluxDB and deploy an IODA Grafana instance. This will include the traditional IODA time series (IBR, BGP and Active Probing) as well as time series from other projects, e.g., Google Transparency Report. We will also experiment with importing OONI data in the form of annotations to be visualized jointly with timeseries data in our graphs. We will then conceive, design, and implement Grafana dashboards to enable correlation of all the available data sources. Our preliminary analysis indicates that this will also require the implementation of few simple custom UI components for Grafana.

Task 2: Improve IODA’s usability: Create Basic/Advanced versions of IODA UIs

One of our core objectives is to support more immediate response (by journalists, lawyers, policy makers, censorship circumvention projects, and members of the Internet freedom community) to global censorship events, to help advance Internet freedom and human rights around the world. In particular, over these years, we have been observing and documenting Internet shutdowns in Africa and the Middle East. This task aims at making IODA more accessible to the Internet freedom community. Our interfaces were originally designed with academic computer science researchers and network operators in mind and they currently reflect the underlying complexity of the system. By making IODA’s interfaces more accessible to a diverse audience—including groups working on developing and deploying censorship circumvention tools—we will improve its effectiveness in empowering the Internet freedom community at large.

With the upsurge in users who are using IODA for outage detection and analysis, we have been learning more about which user interface (UI) features can help users further and started introducing them in IODA. However, we also learned that our public Web dashboard interfaces can be confusing to less technical users. We received specific feedback from multiple parties working closely with activists and civil society indicating that a much simplified version of such interfaces would increase the usefulness and uptake of IODA. To keep the availability of more advanced features and detailed data, we plan to develop dual versions of our dashboards (basic and advanced) and in some cases introduce toggles to enable features for more advanced users without interfering with the experience of novice users. Dual interfaces will allow us to preserve the availability of more advanced features and access to more detailed data to IODA power users.

Task 3: Engage in outreach and collaboration with censorship circumvention and Internet freedom groups and release timely reports about shutdown events

We plan to broaden our collaborations—and strengthen the existing ones—with groups operating in the field of Internet freedom. In particular, we propose to extend and intensify our collaborations with groups developing and deploying Internet censorship evasion tools and strategies, with the goal to better understand and improve how IODA can support decision making, deployment strategies, and development of circumvention efforts. In this task, we plan to intensify our existing collaborations with groups like Kandoo, Tor, and Psiphon to better understand their needs outside of and during shutdowns and in order to orient and prioritize our efforts accordingly. This work will include collaborating with these groups to prepare in advance for when new episodes of connectivity disruption might happen, setting up procedures for information sharing, vantage points in and outside the country, and dashboards.

We disseminate our analyses and inform the Internet freedom community about IODA and its new functionalities through several means. We actively participate in the KeepItOn and OTF-Talk mailing lists and we will continue to participate in the mailing list discussions and provide insights about specific events, references, and tutorials. We also attend and present at events such as the Citizen Lab’s Summer Institute (2018), RightsCon (2019, 2020) and the Internet Freedom Festival (2020). We recently took part in a program (Optima) organized by InterNews to train activists in various African countries in the use of IODA to monitor and investigate Internet shutdowns and we will potentially extend this program to other countries. We plan to attend (virtually, when necessary) additional Internet freedom forums, including the next editions of CLSI, RightsCon and the Internet Freedom Festival.

We will continue to use the IODA Twitter account for timely dissemination of detected outages.. We will also publish detailed analyses providing insight into interesting events either using the CAIDA blog or in collaboration with other groups. We regularly collaborate with OONI on the analysis of shutdown events in several countries (recently in Iran ). Since December 2018 we have (co)authored four blogposts/reports offering in-depth analyses of shutdown events. Consistently with our experience, we expect to (co)author a blogpost once every 3-4 months (depending upon the occurrence of shutdown events). We are also working with third parties that use IODA data and visualizations as supporting evidence for their work. These analyses may result in reports, albeit not co-authored by our team. We will continue to collaborate with other groups operating in this space and foster new collaborations.

Over the last couple of years, the IODA team has been collaborating with international digital rights organizations, such as Access Now and Amnesty International. Since 2018, IODA has been an active member of Access Now’s KeepItOn campaign (an international coalition of human rights defenders fighting internet shutdowns) supporting advocacy efforts with censorship measurement data. We also aim to continue supporting circumvention tool projects, such as the Tor Project and Psiphon , to help them improve strategies on rapid response to emergent censorship events.

Last Modified