Introduction

Based on incidents such as those reported in https://www.ft.com/content/24b8b7b2-9272-11e9-aea1-2b1d33ac3271, there is interest in spotting the occurrence and impact of large Internet outages such as may be occasioned by civil unrest, cable cuts, cable upgrades etc. (see the article "From Content Blocking to National Shutdowns: Understanding Internet Disruptions" from the Internet Society for the various causes of outages). For a list of outages examined by PingER see PingER Case Studies.

Automating detection

To spot the shutdown of a country automatically would take a bit of mining. For example, the country should have 2 or more target hosts and if all the targets in a country go offline (not reachable) in the same time frame then that is an indicator that something is probably happening and worth investigating. Of course, that assumes the Measurement Agent (e.g. SLAC) also was not down (due to maintenance etc.) at the same time. One can tell from say https://www-iepm.slac.stanford.edu/pinger/sites-per-country.html how many targets there are by country.

Outages that make hosts unreachable are relatively easy to detect since they exhibit 100% loss. Nowadays however, there are often backup routes so that the host stays connected even though a route may not be working. In this case, there is usually a step-change in the minimum RTT since the new route has a different distance to the original route. Also, the new route may have different capacity and will be shared by a greater number of conversations, thus it is likely to be more congested than the original route so the jitter and average RTT may change.

All hosts in a given country may not use the same carriers and thus have different routes so one needs to look at multiple hosts in a given country.

Some of the outages are all day(s) which would make it easier to detect.  Shorter outages would be trickier.  On the other hand, if we know of outages then we can look at the data around that time to see the impact.  

The article (https://www.ft.com/content/24b8b7b2-9272-11e9-aea1-2b1d33ac3271) talks about mobile networks that we are not monitoring. It also refers to turning off social media which probably would not affect PingER. Also, there is no kill switch for a country, there are typically multiple carriers who will probably shut down at different times.  

Some Examples

The examples are taken from PingER Case Studies

Cable cuts

In January 2020 there were two cable cuts along the west coast of Africa that affected sites in Southern Africa, see: South Africa damage to the SAT-3 and WACS undersea cable Jan 16 2020. In this case there was no 100% packet loss since the routes automatically switched to alternate-working routes, however, there was a significant step-change in the minimum RTT.

Another example is the impact of the Japanese Earthquake and Tsunami March 11th, 2011. Again there was a significant step-change in the RTTs from SLAC to many Japanese hosts and new routes were chosen to avoid damaged cables to the US.

Also see:

Political events

For a discussion on Internet shutdowns and their impact see https://www.internetsociety.org/policybriefs/internet-shutdowns. There were several political events in 2018-2019  reported in the Financial Times, that impacted Internet performance, see Internet performance and political decisions 2019.
Other examples are:
  • No labels