Internet speed tests: download, upload and latency

An internet speed test measures throughput between your device and a particular test server at a particular time. Most tests show download, upload and latency separately. The result depends on the server, route, local network and current load; one measurement cannot describe the performance of every website or application.

What the numbers mean

Download rate describes how quickly data comes to you, while upload rate describes how quickly you send it. Rates are commonly reported in megabits per second (Mbit/s), not megabytes per second: eight bits make one byte. Latency, often shown as ping or round-trip time, measures the delay for a request and response. Jitter and packet loss can also matter for calls and games. A test does not directly measure the Wi-Fi link rate shown by your device. The FCC broadband report treats download, upload, latency and packet loss as distinct measures.

How to run a useful test

For a comparison with a fixed broadband plan, use a wired computer if possible. Pause large downloads, software updates and cloud backups. Turn off a VPN if you are assessing the service without it; keep it on if the VPN connection is what you want to test. Repeat the measurement at different times and, if needed, with different test servers. Record the connection type, server, time and how many people were using the network. To investigate Wi-Fi, test near the router and again where the problem occurs.

Why results differ

The selected server may have a short, uncongested path while a streaming service or work server uses a different route. The test device, browser, router, Ethernet port, Wi-Fi interference, mobile signal or remote server can limit the result. A computer with a 100 Mbit/s Ethernet port cannot demonstrate a 1 Gbit/s broadband rate over that port. A displayed Wi-Fi link rate is not the amount of useful application data transferred: wireless airtime and protocol overhead reduce it. Different test methods, durations and parallel flows can also produce different figures.

Comparing the result with your plan

Check whether your provider advertises a maximum or a guaranteed rate, in which direction, and under which measurement conditions. A deviation above an arbitrary percentage, such as ten percent, does not on its own prove a fault. If low rates recur over a wired connection and across several servers, save a series of results and contact the provider. If only one service is slow, check its own status and compare other destinations. If calls suffer despite a high throughput result, test latency and loss under load as well as without load.