"I pay for fast fibre but the speed test shows half of it": it is one of the things I hear most often. Sometimes the line really does have a problem. Far more often the test is measuring something else: the Wi-Fi, a slow port, a busy computer or a distant server. A useful speed test is not the one with the highest number, but the one that isolates the line from everything else.
What a speed test measures
A speed test downloads and uploads data from a server for a few seconds and calculates the average speed, as well as the response time. The result depends on every link in the chain:
- the device running the test and its browser or app;
- the connection between device and router, cable or Wi-Fi;
- the router itself and its ports;
- the line to the provider;
- the provider's network and the path to the server;
- the test server and how busy it is.
If any of these links is slower than the line, the test measures that link. That is why you need to eliminate as many of them as possible.
Rules for a clean test
Use a cable, not Wi-Fi
Wi-Fi is the number one cause of low results. Even with an excellent signal, real speed depends on distance, walls, neighbouring networks and the device's capabilities. To measure the line, connect a computer to the router with an Ethernet cable. A Wi-Fi test is still useful, but it measures the Wi-Fi.
Check the port speed
A 100-megabit Ethernet port, a cheap USB adapter or a faulty cable cap the result at about 94 Mbit/s, a typical value that is easy to recognise. The computer's network settings show the negotiated link speed: for a line faster than 100 megabits it must be at least 1 Gbit/s. Lines above a gigabit need multi-gigabit ports and cables along the whole path.
Free up the network and the device
During the test other devices may be downloading updates, syncing photos or streaming video. Pause backups, close other browser tabs and, if possible, run the test when the house is quiet. An active VPN on the computer also changes the result completely.
Choose the server sensibly
Most services pick a nearby server automatically. It is worth trying two or three different ones: an overloaded server, or one poorly connected to your provider, gives low results that have nothing to do with you. Some providers offer their own test server, which is especially useful when you open a complaint.
Repeat at different times
A single test is a snapshot. Three tests in the morning, three in the afternoon and three in the evening, written down, show whether there is a regular drop at peak times, which points to congestion in the provider's network or the server's.
Reading the values
| Value | What it indicates |
|---|---|
| Download | How fast you receive data: streaming, downloads, browsing |
| Upload | How fast you send data: video calls, backups, uploads |
| Idle latency | Round-trip time on an unloaded line |
| Latency under load | Response time while the line is full, a sign of bufferbloat |
| Jitter | Variability of latency, important for voice and gaming |
The advertised speed in the contract is almost always the theoretical maximum of the technology. In practice part of the bandwidth goes to protocol headers: on a gigabit line it is normal to see values a little below 1000 Mbit/s even when everything is working.
Testing from the router
Many routers include a speed test in their interface. It has the advantage of removing the device and the internal network, but the drawback that the router's CPU, which is also busy routing traffic, may not be enough to measure very fast lines. If the router test and the cabled test agree, the result is reliable. If they differ widely, one of them is limited by hardware.
When to report it to the provider
Before opening a complaint, gather the data: cabled results, confirmed 1 Gbit/s port, different times of day, several servers, with date and time. In my work I have found that a table like this is taken far more seriously than a generic "the line is slow", and it immediately shows whether the problem is constant or tied to certain hours.
In short
- A speed test measures the slowest link in the chain, not just the line.
- To measure the line, use a computer cabled to a gigabit port.
- A result stuck around 94 Mbit/s points to a 100-megabit port or cable.
- Repeat tests with different servers and times and write the results down.
- Look at upload, latency under load and jitter too, not only download.
