Vision Statement
Despite almost 30 years of research in devising various methods to measure network capacity and throughput, the speedtest by Ookla—the most popular and global speed test platform measurement — resorts to the non-novel flooding method to estimate the throughput between a speed test client and a configurable speed test server. The browser-based Ookla speed test employs a single TCP connection to measure the bulk transfer capacity and parallel connections the achievable TCP throughput on the HTTP level. Notwithstanding the simplicity of the measurement methods, many nondisclosed details about how to compute the single download/upload throughput value obscure what the reported results actually measure.
Our research will aim to demonstrate that a number of very different throughput results can be computed from the same measurement dataset obtained from a browser-based speed test client. There are multiple definitions related to throughput measurement, and it is unclear which one(s) Ookla actually measures. It can be argued that such a precise definition is perhaps not needed for most users. However, we counter-argue that a more precise understanding is called for as these data are increasingly used for academic research, decision making, and addressing digital equality.
Project Background
This project was initially started in 2023 by CAIDA at UC San Diego, where they developed an open source toolkit called RABBITS. This project was quite different from the current scope today. This toolkit was a wrapper over the six most popular Internet speed test platforms, and was designed to standardize parameters across the platforms. Further information about RABBITS can be found here. Anwesha Pradhananga and Aishwarya Joshi began working on this project in the Spring of 2024 to extract important Netlog data captured during the test and perform an initial analysis of RABBITS using the Ookla speed test server. In their analysis, Anwesha and Aishwarya examined the effect of the number of TCP flows and the size of the HTTP message request on the throughput measurement. I took over this project in the Spring 2025 semester.
When I began this project, I was working on the RABBITs toolkit but transitioned to investigating the underlying implementation of Ookla specifically. By investigating Ookla in more detail, the goal was to return to RABBITs to validate its behavior and perform a comparative analysis. I produced a paper for my final in the class, in which the findings have served as foundation for the work we have done this year. Over the Summer of 2025, Joshua Park worked on this project as a summer researcher for Calvin.
Measuring the internet has a substantial community, and there is an annual Internet Measurement Conference (IMC). The IMC allows papers to be published on various internet measuring investigations. ArXiv also has a repository of related research to internet measuring, such as this paper on early termination of speed tests to reduce unnecessary traffic. The paper has a submission deadline of April 30, 2026 at 8am.
Most Internet speed test platforms, except for M-Lab, generally use the same measurement method which is to flood the network with synthetic data to the point where a bottleneck is reached. This bottleneck is the narrowest bandwidth in both the download paths and upload paths. In order to flood the bottleneck along the network path, a large amount of data must be sent. If the volume of data is too low, the bottleneck will not be saturated and throughput measurements will be an under-estimate of the true throughput. Speed tests are also traditionally performed on the “last mile” server, which is the closest server to the user.
Team Members
- Ben Kosters: A senior at Calvin University who is focused on learning more about networking and security.
- Priscilla Chen (CS390): A junior at Calvin University who is focused on learning more about data analysis.
Project Advisors
- Professor Rocky Chang
- Dr. Ricky Mok (CAIDA)
- Tanmay Nale (UCSD)