Broadband speed test

How it works

The tester opens a path to a measurement server and moves data in both directions. Download is how quickly the server can fill this device. Upload is how quickly this device can send. Delay, often labelled ping or latency, is how long a small packet takes to go and return. Some testers also report jitter, the jumpiness of that delay, and packet loss, the data that never arrived. A complete reading names those axes. A single megabit headline is not a complete reading.

Broadband here means an always-on home or office link: fibre, cable, DSL, fixed wireless, or a similar last-mile path. The test does not know your contract. It knows this radio or this cable, this moment, and the server it reached. A run through a phone hotspot is a cellular test wearing a broadband label. A run through a guest network with a cap is a capped test. Name the path you meant before you quote the number.

The server location is part of the instrument. A nearby server times a short path. A distant server adds hops that are not “the broadband in this room.” Evening congestion on a shared node can sag a result that looked fine at noon. One screenshot is a moment. Two or three runs a few minutes apart are a pattern. If the three disagree wildly, you measured load, a weak radio, or a busy device, not a single honest speed.

Download: data arriving, usually in megabits per second. Upload: data leaving, same unit. Latency: round-trip delay, usually in milliseconds. Jitter and loss: how stable that delay was, and whether packets vanished.

Give the hose a fair test

Pause the other jobs that drink the same pipe. Cloud backups, game updates, video streams, and a second person on the same Wi-Fi all sit in the reading. The test is still real. It is no longer a reading of an idle line. If you came here to judge the broadband, idle the house, then run. If you came here to judge the busy house, leave the traffic on and say so on the note you keep.

Wi-Fi is air, walls, and a band choice. Ethernet is a cable. Those are different instruments. If the cable looks fine and the air looks poor, the access point or the room is the suspect, not the street. If the cable looks poor too, look at the gateway, the line, and whether the test server is far away. Do not replace a router to fix a distant server, and do not stand in the garden to fix a bad patch lead.

Join the network you mean to judge, not a neighbour’s guest SSID. Turn a VPN off unless the tunnel is the thing you meant to time. Leave the test tab in the foreground until the run ends. Write download, upload, delay, wired or wireless, and the time of day.

Repeat once. Keep both snapshots. Phones and computers on the same Wi-Fi often disagree because antennas, radios, and background updates disagree. Compare like with like: same device, same room, same band if you can choose it. A 5 GHz run and a 2.4 GHz run are not twins. Treat them as two tests.

Megabits measure the hose. Megabytes measure the bucket. Mixing them is how a “fast” line still copies a file slowly.

What the numbers are for, and what they are not

Download is the figure people quote because video, updates, and incoming files live there. Upload is the figure that makes a camera freeze on a call while a film still plays. Delay is the figure that makes a game or a voice call feel late even when the megabit numbers look generous. A wide hose with a sleepy gate is still a sleepy gate.

Units are not decoration. Eight megabits are one megabyte. File sizes on disk are usually bytes. Speed tests are usually bits. Do not add ping to download; they are different axes. Do not treat a plan’s “up to” ceiling as a floor the test failed to meet. The plan is a cap. The test is a sample under the cap, through this path, in this minute.

When you talk to a provider, take the three numbers, the device, wired or wireless, and the hour. A complaint without those details asks someone else to guess what you saw. The broadband speed test’s job ended at the snapshot. Diagnosis starts with whether you measured the path you thought you measured.