NeTS: SMALL: Broadband Service Reliability: Characterization And Improvement

While providing access and sufficient capacity remains a challenge in a great part of the world, in many developed countries broadband services continue to improve rapidly propelled by government and private interests. With widely available and faster broadband, consumers start to migrate key services such as television, telephony and home monitoring to over-the-top alternatives, and their expectations regarding broadband reliability are becoming paramount, ranking first in recent reports on consumer experience. Yet, despite the rich history of reliability engineering we have very limited understanding of reliability in the context of broadband, from the most appropriate metrics for characterizing it, to approaches for gathering measurements or assessing its impact on quality of experience. When reliability problems appear, troubleshooting them is nearly impossible—there is no principled approach to determining whether the source of the problem is the user’s device, DNS, the home wireless, the access link, the Internet service provider, or anything else in the network. The outcome is increased frustration for users, growing management costs for Internet service providers, and, ultimately, underperforming applications. We study the reliability of broadband services. Our research agenda has three parts: (1) Characterizing broadband service reliability: We will develop techniques and explore metrics to study and characterize reliability problems in broadband services, and algorithms to diagnose and localize common problems. To perform this study, we will build on our existing research infrastructure - the namehelp and Dasu platforms - and the data collected by the FCC Measuring Broadband America effort. (2) Understanding the relation between reliability and QoE: We will explore the impact of reliability problems on users’ QoE with different services, investigating the value of possible proxy metrics for QoE (e.g., session time, traffic demand) and the application of natural experiments and related designs to understand their relationship at scale while controlling for most confounding factors. (3) Exploring edge-based solutions to improve reliability. With an understanding of reliability problems, impact and most common causes, we will investigate possible solution that might improve service reliability leveraging our platform deployments in access networks and exploring new middlebox-based solutions.