Pingin’ in the Rain

If you live at the end of the line for an internet connection, you may also be at the end of your rope.

Severe weather doesn’t just pose a significant threat to the stability of the nation’s economic and digital infrastructure—in the so-called “last mile” internet, it can deliver a knockout blow to your Netflix, axe your Amazon purchases and foul up your Facebooking. While we know how to calculate storm damage and casualties, weather-related disturbances in this final leg of connectivity to customers was long a mystery.

Now a team of computer scientists at the University of Maryland is zeroing in on the problem with software to quantify the extent and duration of internet outages observed during weather-related events.

Assistant Professor Dave Levin and Professor Neil Spring recently presented the study on the effects of weather events—including seasonal rain and snow—as they relate to residential internet outages during the 2019 meeting of the Association for Computer Machining’s Special Interest Group on Data Communication. Both researchers have appointments in the University of Maryland Institute for Advanced Computer Studies.

The team also included lead author Ramakrishna Padmanabhan M.S.’17,  Ph.D. ’19 and Aaron Schulman Ph.D. ’13.

“Residential links are often our primary means of connecting to the internet, and these links are continuously exposed to different weather conditions,” said Padmanabhan.

The UMD team found that investigating internet outages in residential networks due to weather was challenging due to several factors—including media types, internet protocols and different service providers—all in varying contexts of climates and geography.

To address these issues, the researchers collected eight years of data on active outages across the bulk of the last-mile internet infrastructure in the U.S. Their “ThunderPing” software tracks weather forecasts in the U.S., then pings a sample of 100 internet customers from each last-mile provider before, during and after a weather event to check the status of their service.

After analyzing 1,811 days of data—representing 101 billion responsive pings to 10.8 million residential addresses across a range of seasonal and regional weather conditions—they showed how a variety of factors influence the likelihood of internet dropouts, including type of weather, link type and geographic location.

Team members said their study can benefit multiple stakeholders: Businesses can identify the extent to which they lose customers to commonplace weather events; governments can identify areas that are particularly vulnerable to weather-related outages and promote investment in alternate technologies; and customers can learn which providers in their areas are resilient to different types of weather conditions and make informed choices.

The research was funded by National Science Foundation’s Division of Computer and Network Systems grants, the U.S. Department of Homeland Security Science and Technology Directorate and the Air Force Research Laboratory.