ISP access networks are critical to the Internet. They aggregate millions of users through diverse infrastructure (e.g. cellular base stations and fiber), and route traffic through central offices to connect to the Internet backbone. Although access networks are fundamental to Internet connectivity, their design is not standardized. Each ISP builds its access networks differently with different vendors and designs across regions, and multiple ISPs deploy distinct access networks within the same region to compete for customers. These variations can lead to differences in access network performance and resilience across providers and regions.

However, accurately measuring access network performance and resilience is a challenge for end-users, regulators and even ISPs themselves. Several problems contributes to this difficulty. First, regional access network infrastructure often blocks external probes to prioritize customer traffic, limiting active measurements. Second, measuring wireless last-mile infrastructure requires close proximity to capture signals above the noise floor. Third, field measurements are not always reliable—ordinary customers using the service and environmental noise can significantly affect performance analysis. These problem arises from diverse components in access networks and region-specific deployment strategies. Network measurement approaches that rely on limited vantage points and inference techniques, can fail to capture regional variations comprehensively.

In this dissertation, I introduce new methodologies to overcome these challenges by using public resources to create vantage points in different regions. Through extensive field measurements, I reveal that ISPs’ access networks exhibit highly diverse designs across regions, providers, and vendors. I evaluate how these design disparities impact latency, traffic performance, and network resilience. My contributions provide a foundation for improving the transparency of regional access networks and evaluating their resilience and performance.

In summary, I defend the following thesis statement: The diversity and opacity of regional access network infrastructure hinders accurate evaluation of its performance and resilience (e.g. network outage time, downlink throughput, wireless SNR), which can be addressed through: (1) Using local public Wi-Fi networks and public transit mobile phones to reveal access network topology and assess physical risks including the root cause of access network outage, (2) using controlled mobile phone experiments to uncover LTE base station scheduler design variations across vendors and evaluate the impact on downlink throughput, and (3) integrating mobile devices with software-defined radios to evaluate wireless signals under a variety of interference in the field.