TURBOTEST: Learning When Less is Enough through Early Termination of Internet Speed Tests
This addresses the growing traffic costs for users, ISPs, and policymakers by making speed tests more efficient, though it is an incremental improvement over existing platforms.
The paper tackles the problem of reducing data usage in internet speed tests by deciding when to stop tests early without losing accuracy, and introduces TURBOTEST, which achieves 2-4x higher data savings than existing methods while reducing median error.
Internet speed tests are indispensable for users, ISPs, and policymakers, but their static flooding-based design imposes growing costs: a single high-speed test can transfer hundreds of megabytes, and collectively, platforms like Ookla, M-Lab, and Fast.com generate petabytes of traffic each month. Reducing this burden requires deciding when a test can be stopped early without sacrificing accuracy. We frame this as an optimal stopping problem and show that existing heuristics-static thresholds, BBR pipe-full signals, or throughput stability rules from Fast.com and FastBTS-capture only a narrow portion of the achievable accuracy-savings trade-off. This paper introduces TURBOTEST, a systematic framework for speed test termination that sits atop existing platforms. The key idea is to decouple throughput prediction (Stage 1) from test termination (Stage 2): Stage 1 trains a regressor to estimate final throughput from partial measurements, while Stage 2 trains a classifier to decide when sufficient evidence has accumulated to stop. Leveraging richer transport-level features (RTT, retransmissions, congestion window) alongside throughput, TURBOTEST exposes a single tunable parameter for accuracy tolerance and includes a fallback mechanism for high-variability cases. Evaluation on 173,000 M-Lab NDT speed tests (2024-2025) shows that TURBOTEST achieves nearly 2-4x higher data savings than an approach based on BBR signals while reducing median error. These results demonstrate that adaptive ML-based termination can deliver accurate, efficient, and deployable speed tests at scale.