Pingin' in the Rain

Observing how weather affects residential Internet connections

The goal of this work is to investigate the resilience of residential Internet connections before, during, and after routine and severe weather events. When the National Weather Service issues a weather alert for a county in the United States, we find IP addresses belonging to residential Internet connections in that county. We ping each host once every eleven minutes from ten different PlanetLab hosts, beginning up to six hours before the alert takes effect and ending six hours after it expires.

Datasets

Our Thunderping datasets are available for download.

Papers

Analysis of weather's effect on Internet outages

Residential Links Under the Weather
Ramakrishna Padmanabhan, Aaron Schulman, Dave Levin, and Neil Spring
ACM SIGCOMM 2019

Pingin' in the Rain
Aaron Schulman and Neil Spring
IMC 2011 (Internet Measurement Conference) - Short paper

Techniques we have developed to perform our analyses

How to find correlated Internet failures
Ramakrishna Padmanabhan, Aaron Schulman, Alberto Dainotti, Dave Levin, and Neil Spring
PAM 2019 (Passive and Active Measurement Conference)

Analyzing Internet Reliability with Probing-based Techniques
Ramakrishna Padmanabhan
Ph.D. Dissertation, 2018

Advancing the Art of Internet Edge Outage Detection
Philipp Richter, Ramakrishna Padmanabhan, Neil Spring, Arthur Berger, and David Clark
IMC 2018 (Internet Measurement Conference) - Long paper

Reasons Dynamic Addresses Change
Ramakrishna Padmanabhan, Amogh Dhamdhere, Emile Aben, kc claffy, and Neil Spring
IMC 2016 (Internet Measurement Conference) - Long paper

Timeouts: Beware Surprisingly High Delay
Ramakrishna Padmanabhan, Patrick Owen, Aaron Schulman, and Neil Spring
IMC 2015 (Internet Measurement Conference) - Long paper

People

UMD

Aaron Schulman
Ramakrishna Padmanabhan
Youndo Lee
Patrick Owen
Dave Levin
Neil Spring

Collaborators

Emile Aben
Arthur Berger
kc claffy
David Clark
Alberto Dainotti
Amogh Dhamdhere
Philipp Richter

Severe Events

Hurricane Sandy (2012)
This animation shows the effect of Hurricane Sandy on the 70,000 hosts we pinged on the US eastern seaboard. A green or red dot indicates the responsiveness of each host. Green means the host is responding to our pings (up) and red means the host is not (down). When a host transitions from up to down a larger dot appears over the host to draw your eye to the locations where hosts are going down. For each state, the table on the right lists the percentage of hosts that are up (%UP) and the number of hosts (#HOSTS) we were pinging.

Hurricane Irene (2011)
This animation shows the effect of Hurricane Irene on the 80,000 residential hosts we pinged in the US. The transparent circles indicate the responsiveness of the hosts. The UP and DOWN states are the same as the Sandy animation. In this animation the HOSED state indicates the host is only responding to some of our pings. The diamonds indicate weather observations at various airports across the US. The video ends abruptly when we lost power at the University of Maryland.

If you would like the data that we collected for this work please contact nspring@cs.umd.edu.