Measuring Internet access network performance has been a persistent challenge for researchers and policymakers alike. Unfortunately, existing “speed test” datasets typically lack comprehensive data across both space and time. Specifically, our past work has highlighted that tools like Ookla’s Speed Test and Measurement Lab’s NDT rely heavily on convenience samples (user-initiated tests from self-selected participants), resulting in a sample that may not generalize across either time or geography. Our ongoing research seeks to address these issues by developing innovative sampling methods and statistical models to provide a more holistic view of Inter- net performance. Initial findings, focusing on end-to-end la- tency across hyper-local regions within a single large city in the United States (Chicago, Illinois), reveal that spatial proximity often does not correlate with simultaneous perfor- mance anomalies. These insights underscore the need for advanced methods to generalize Internet performance data across time and space. Improved methods can ultimately en- able a better understanding of the effects of infrastructure investments on the evolution of Internet performance.
May 05, 2022