Internet Speed Test

How does the speed test measure download, upload and latency?

Three separate measurements run in sequence, all from your browser. The first step measures latency: small packets bounce between browser and server, the average round-trip-time gives the ping in milliseconds. The variance of those values is the jitter – how consistent the connection is. For real-time communication, consistency often matters more than the average latency itself.

The second step runs the download test: the browser requests packets of varying sizes and measures how many bytes per second actually arrive. The procedure starts with small packets (1 MB) and ramps up to larger ones (25 MB), so it captures both latency-dominated short bursts and bandwidth-dominated long transfers. The 90th percentile is taken across multiple runs – that filters out brief dropouts and yields a realistic maximum.

The third step does the same for upload, except packets are sent from browser to server. While download and upload are running, latency under load is measured in parallel – the value that drives the bufferbloat grade. Source for the methodology: the IETF spec for network quality measurement, which Apple has been working on since 2022.

What do the numbers mean for your day?

Most speed tests throw four numbers at you and leave you to translate. Not here – below the headline block sit four concrete use-case statements, computed directly from your measurement. The rules of thumb the translator uses:

• 4K Netflix stream: about 25 Mbps download per stream. With 100 Mbps you can run four parallel 4K streams without buffering.
• HD video call: about 5 Mbps down and 4 Mbps up per device. With 25 Mbps down and 10 Mbps up, five parallel HD calls work.
• Game patch download: a 50 GB patch takes about 70 minutes at 100 Mbps download, only 14 minutes at 500 Mbps. Formula: GB × 8000 / Mbps = seconds.
• Live streaming upload: 6 Mbps upload is enough for 1080p Twitch or YouTube live, 16 to 25 Mbps for 4K live upload. Anyone trying to stream with less than 3 Mbps upload ends up at 480p with compromises.
• Cloud backup upload: Mbps × 3600 / 8000 = GB per hour. At 30 Mbps upload that is 13.5 GB per hour.

These numbers are deliberately conservative. Real streaming services have adaptive bitrate management that lowers stream quality on buffering – the rules of thumb show what you can do smoothly, not the absolute maximum.

What is bufferbloat — and why is the grade more important than the ping?

Bufferbloat is the phenomenon where your latency rises dramatically under load. A connection can sit at 12 ms ping while idle, then climb to 250 ms under download load – the classic bufferbloat-F pattern. Consequence: a voice call that was fine before starts to stutter as soon as someone in the same network downloads a big update. Online gaming becomes a lottery. Video calls show the classic robot effect.

The cause sits in oversized packet buffers in routers and ISP equipment: under load, they fill with packets that get stuck in queue order. Latency rises because every new packet has to line up behind. Quality of Service on the router (Smart Queue Management, fq_codel, CAKE) solves the problem, but is rarely activated on stock hardware.

The grade A through F is based on the delta between idle ping and loaded ping:

• A = delta ≤ 5 ms: even online gaming stays stable under load.
• B = 5 to 30 ms: video calls work reliably.
• C = 30 to 60 ms: calls may stutter when someone else downloads in parallel.
• D = 60 to 200 ms: live gaming and calls suffer during any background download.
• F = > 200 ms: connection becomes unusable once anyone uploads.

Anyone who lands a bad grade can often improve significantly with Smart Queue Management on the router (DD-WRT, OpenWrt, some stock firmwares). On stock devices without that option, a modem swap from the ISP sometimes helps.

Does the measured value match your plan?

If you filled in the plan field above, the tool computes the plan match in percent and slots it into a tolerance band:

• ≥ 95 percent: Full performance. You get what you pay for.
• 80 to 95 percent: In the normal wired range. ISPs rarely deliver exactly 100 percent.
• 60 to 80 percent: Normal for WiFi, below the wired expectation band.
• 50 to 60 percent: Below the typical range. Try restarting the router, swapping the cable, or calling your ISP.
• < 50 percent: Well below your plan. Contact ISP support.

Worth knowing: ISPs often sell “up to X Mbps”, which in the small print means 80 to 95 percent of that is the regular range. Source: Allconnect on advertised vs actual speeds. Anyone who booked 250 Mbps and measures 215 Mbps (86 percent) is getting what they paid for. Anyone who booked 250 Mbps and measures 80 Mbps (32 percent) has a real problem.

Why do WiFi and cable results differ so much?

The most common complaint in any tech-support thread: “My 1 Gbps plan only delivers 200 Mbps.” In nine out of ten cases, that is not the ISP, it is the WiFi. Three factors play in:

• Radio range: the signal halves with every wall. 600 Mbps in the room with the router, 80 Mbps three rooms away – both measured on the same plan.
• Frequency band: 2.4 GHz reaches further but is slower and full of competition from neighbouring networks, Bluetooth devices and microwaves. 5 GHz is faster but the range is significantly shorter. Wi-Fi 6E and Wi-Fi 7 use the 6 GHz band, which is almost empty – correspondingly fast, but even shorter in range.
• Concurrent devices: every connected device shares air time with the others. A smart TV streaming, a few IoT devices and a phone all run in parallel.

Practical: for an honest plan verdict, test at least once with an Ethernet cable directly into the router. That is the number your ISP promised – everything in between is on your local WiFi setup, not the provider.

What is the test useful for?

Preparing an ISP complaint. When the internet feels slow, the test gives you a concrete number plus a plan-match percentage. ISPs respond differently to “I am only getting 28 percent of my plan” than to “internet is slow”.

Before switching plans. Anyone who just moved into a new place and is wondering whether 100 or 250 Mbps is enough: a speed test with current values plus the use-case statements (“4 parallel 4K streams”) gives a tangible decision basis.

Evaluating router optimisation. Anyone who connected a new router or set up Quality of Service wants to know whether it helped. Before-after comparison, ideally tested multiple times.

Bufferbloat diagnosis. Anyone with stuttering on voice calls or online games: a bufferbloat grade D or F explains the problem directly and suggests what to do (enable Smart Queue Management, swap the router, evaluate the ISP modem).

During a network outage. When the internet suddenly stops working: a test quickly shows whether it is a bandwidth problem (all values near zero), a latency problem (values normal but ping > 500 ms), or nothing (test fails completely). Valuable for the ISP phone call.

Which tools pair with this?

From the kittokit dev-workflow ecosystem:

• Hash generator — SHA-256, MD5 and friends for file-integrity checks after downloads.
• JWT decoder — decode JSON Web Tokens, useful when debugging API authentication over the connection.
• UUID generator — generate unique IDs, often needed when setting up new API endpoints or test data.