Speed Test Methodology

HanoiLUG's speed test uses the NDT7 protocol from M-Lab, the open testing platform run by Google, the New America Open Technology Institute, and Princeton University. NDT7 is reproducible, the server and client code are open source, and the entire measurement dataset is public. That matters because most popular speed tests run measurements through proprietary code against servers whose locations and configurations cannot be independently verified.

Why NDT7

NDT7 uses standard TCP with the BBR congestion control algorithm and tests via a single stream, the way a real application transfers data. Many commercial speed tests open multiple parallel streams to inflate the headline number — that produces a result that looks better than what your browser, your video call, or your downloader will actually see. The single-stream design is the honest version.

What the test does

  1. Connects your browser to the nearest M-Lab server via a WebSocket session, selected automatically based on routing distance.
  2. Downloads data for approximately 10 seconds, sampling throughput and TCP-level diagnostics (cwnd, RTT, retransmissions) every 250ms.
  3. Uploads data for approximately 10 seconds with the same instrumentation in the reverse direction.
  4. Reports the median throughput across the measurement window, discarding the slow-start ramp-up and the final tear-down so the result reflects steady-state performance.

Limitations to read your result against

  • Server proximity. Results depend on which M-Lab server you reach. If the nearest one is geographically distant from you, both latency and achievable throughput will look worse than your connection's actual capacity to a nearer server.
  • End-to-end, not just ISP. The test cannot isolate ISP issues from your Wi-Fi, your router, your device's network stack, or congestion on your local segment. A wired test from a client directly attached to the router is the cleanest signal.
  • Mobile network variability. On cellular connections, tower load changes minute-to-minute. Re-running the test several minutes apart can produce materially different numbers — this is real performance variability, not measurement noise.
  • Asymmetric plans. Most home plans are asymmetric (download > upload). A "500/50" plan caps upload at 50 Mbps regardless of physical line capacity; the upload number being lower than download is expected.

How to interpret your numbers

The results explained guide breaks down what counts as good for each number against typical usage, and walks through diagnostic steps when results are below your ISP plan.