NeTS: Small: Residential Network Outage Detection

University Of Maryland, College Park, College Park MD

Investigators

Abstract

Residential networks lack empirical information about service outages and reliability. The absence of clear information, coupled with the relative inexperience of ordinary users combines to hide the sources of connectivity and performance problems. The goal of this proposal is to provide external measurement of outages in residential Internet connections, empirically determining the reliability of different providers and technologies, and recording outage information suitable for diagnosis. Although some projects instrument home networks from within the home, this work looks from the outside, searching for rare events across many more networks and countries for comparisons. This work addresses two key challenges in measuring network reliability from the outside. The first important problem is to determine whether address changes are actually the cause for a network device becoming unreachable to external measurement. The second key problem is in separating individual failures from other failures. For example, with an approaching storm, many users may disable their equipment to avoid lightning damage, but with a passing storm, an outage may affect many users at once. Engineering the measurement infrastructure to adapt to these events and identify simultaneous outages will permit classifying outages. Reliable networks are particularly important as users transition away from dedicated landlines toward voice-over-IP, while still needing the ability to make emergency calls, particularly during events that challenge networks such as earthquakes and severe weather. Predicting reliability permits informed decision-making about migrating to new network technologies--concretely, whether removing conventional wired lines is safe. Datasets from this research will be public and we will maintain a website that presents Internet outage data in a manner that is accessible by the general public.