Project Summary

The latest COVID-19 pandemic dramatically shifted the mode of in-person jobs and traditional schooling to remote working and online education. Teleconferencing tools (e.g., Zoom, Google Hangout Meet, Microsoft Teams) and learning management platforms (e.g., Canvas and Desire2Learn) play important roles in facilitating communication between co-workers, and delivery of course materials and video lectures. These tools heavily rely on cloud infrastructures to transport high-bandwidth video and audio flows. Empirically grounded understanding of the health and performance of Internet infrastructures – including how ISPs connect to each other and to the cloud – is critical for enabling and maintaining effective telecommuting and tele-education.

Scientific measurement of Internet performance is persistently challenging. Over the past decade, researchers and commercial companies have deployed infrastructures to measure network throughput as a performancemetric. Web-based speed tests, e.g., Ookla, are popular among end-users to diagnose slow broadband connections, and to validate whether they can achieve the ISPs advertised speed. Other third-party test platforms, such as M-Lab and Speedof.me, use servers connected to IXPs and Content Delivery Networks (CDNs). Some ISPs (e.g., Comcast, Verizon, and Google Fiber) and content providers (e.g., Netflix) host their own speed test platforms to facilitate measurements for their customers. Unfortunately, these web-based tests only measure throughput at the edge or over a small segment of the path. In particular, these measurements do not represent the actual quality of service of cloud-based applications.

We propose to fill this gap by orchestrating multiple speed tests across the now-critical paths – from the cloud to test servers inside access ISPs. We will dissect our initial tests to obtain lists of available speed test servers in the US.We will then execute speed tests from cloud instances using automated headless browser scripts.