How internet speed is measured

What a speed test does

1. Picks a server. Usually one nearby, to keep distance from dominating the result.
2. Measures latency. It sends small messages and times the round trip. That is the ping, in milliseconds. Many tests also measure it again while the line is busy ("loaded latency").
3. Downloads for a few seconds. It pulls data as fast as the path allows, often over several connections at once so that one slow connection doesn't hold the result back. Some tests deliberately use a single connection instead, which gives lower but different information.
4. Uploads for a few seconds. The same in the other direction.
5. Reports throughput. Data moved divided by time, converted to Mbps. Most tests leave out the first moments while the connection ramps up.

The underlying sum is the same one the network throughput calculator uses: 2.5 GB moved in 30 seconds is 20,000 megabits ÷ 30 = 666.7 Mbps.

Line rate vs what a test can see

A test measures useful data, sometimes called goodput. The physical link carries more than that: every packet has headers, and Ethernet adds framing and gaps between frames. On a 1 Gb Ethernet link, a 1,500-byte packet carries 1,448 bytes of data but uses 1,538 bytes of wire time:

1,000 × 1,448 ÷ 1,538 ≈ 941 Mbps

That is why a working gigabit connection typically tests around 940 Mbps on a cable. Every link type has some overhead of this kind, so a result a few percent below the plan is expected, not a fault.

Why two tests rarely agree

• Different servers. A server further away, or a busy one, gives a lower result.
• Different methods. Tests differ in how many connections they open, how long they run and which samples they discard.
• Your own network. Wi-Fi signal, an older router, the device's network adapter and other devices streaming or updating all reduce what the test sees.
• Time of day. Provider networks are busiest in the evening. A result that drops sharply every evening points to congestion.
• Units. Some apps report MB/s instead of Mbps; multiply by 8 before comparing (the MB/s to Mbps converter does it).

None of these means a test is wrong. Each answers "how fast was this path, just now?", and the path changes.

Latency, jitter and loaded latency

Ping is the round-trip time for a small message, in milliseconds. Jitter is how much it varies from one measurement to the next. Loaded latency is ping measured while the download or upload phase fills the line; if it is far higher than idle ping, queues in the router or modem are adding delay, which hurts calls and games more than a lower Mbps figure would. Latency vs bandwidth covers what each number means and what to aim for.

How providers and regulators measure

Plans are sold as "up to" figures. In the US, providers must also publish typical download and upload speeds and typical latency on FCC broadband labels, which gives a fairer comparison than the headline. The FCC's Measuring Broadband America programme measures delivered speeds across providers with dedicated test equipment in volunteers' homes. For engineers, IETF RFC 6349 sets out a framework for testing TCP throughput properly, including accounting for latency and overhead.

How to test fairly at home

1. Use a computer on an Ethernet cable, close to the router.
2. Pause other downloads, backups and streams.
3. Run two different tests, three times each, and take the typical result.
4. Repeat in the morning and the evening.
5. Compare the typical wired result with your plan in the speed test checker. About 85% or more on a cable is on target by our rule of thumb.

If you want to know what your result means for everyday use, what is Mbps translates plan speeds into streams and download times.