Skip to content
BenchDevice

Storage speed numbers in context

Sequential and random speed, MB/s against IOPS, what a browser-based test measures and how it compares to a native one.

A drive's box carries one big number, and a storage test gives you four smaller ones. Neither is wrong; they measure different things. This guide explains what the four figures on the storage speed test mean, why they differ from each other and from the box, and what a browser can and cannot tell you about the drive underneath.

Sequential and random

Sequential work reads or writes one long stretch of a file from start to end. Copying a video or installing a game is sequential: the drive hands over block after block and can stream them. The test does this with a file of the size you choose, written and read in 1 MiB blocks, and reports the result in MB/s.

Random work jumps about. Booting, opening a browser profile or loading a game level sends thousands of small reads and writes to scattered places on the drive, and each must be found before it is served. The test does this too: 4,096 operations of 4 KiB each, at places chosen by a seeded generator so that every run and every browser hits the same pattern, over a separate 64 MiB file.

Most drives are far faster at sequential work than at random, and the gap is what you feel day to day. A slow sequential figure makes copies take long. A slow random figure makes everything feel late, because the small reads that open programs and files wait in a queue.

MB/s and IOPS

Megabytes a second measures how much data moved. The test counts bytes ÷ seconds ÷ 10⁶, the decimal megabyte drive makers use, so its MB/s line up with the numbers on a box rather than with a file manager's MiB.

IOPS, input and output operations a second, measures how many separate requests the storage answered. The test counts operations ÷ seconds. For big blocks the two tell the same story, so the test shows only MB/s. For small random blocks they diverge: a drive can move little data yet answer many requests, which is exactly what a responsive system needs, so the test shows both.

Each figure you see is the median of three runs. A run that recorded no time is left out, and if none remain the figure stays blank. The median keeps one slow run, from a background task, from standing in for the drive.

What a browser-based test measures

A web page cannot open the drive. What it can open is a private file system the browser keeps for this site, on the same drive, and that is where the test writes. A worker takes a synchronous handle on a file in its own folder, runs the sequential write and a flush, the read, the random writes and a flush, then the random reads, three times over, and deletes the folder when it finishes, when you press Stop, when the page closes, or on your next visit. Before it starts, it checks that the browser has twice its files' size free, and says how much it needs if not.

Between the worker and the drive sit the browser's storage layer, the file system and the operating system's cache. The cache matters most: the system keeps recently written files in memory, so a read that follows a write may be answered from memory and look faster than the drive could ever be. The test flushes its writes so they reach the drive, but it has no way to tell the system to forget a file before reading it back.

How it compares to a native test

A native benchmark talks to the drive with as little in between as the system allows. It can bypass the cache, pick its own queue depth, and write to raw space rather than a file in a folder. Its sequential figures are the ones a box quotes.

The browser test cannot do these things, so expect its figures to differ: often lower, because every operation passes through more layers, and sometimes higher for reads, because of the cache. Treat them as a measurement of the browser's storage path on this computer. That path is the one every web app you use takes, so the figures are real; they are just not the drive's own.

A browser that has no synchronous handles cannot run the test at all, and the page says so. One that has them but refuses them to this page, as private windows often do, gets a different message and no wait.

What to do with the figures

Compare like with like. Run the test twice on a quiet machine and expect the figures to agree; run it on two computers in the same browser and the difference is meaningful. Do not set a browser figure next to the number on a box and call the drive slow.

If the random figures sit far below the sequential ones, that is the shape of normal. If both are low and the machine feels slow, close other programs and run again; if they stay low, run a native test to see whether the drive or the path is at fault. Pick a bigger file size when you want the cache to matter less, and remember that the test deletes its file when it ends.

All guidesBack to the system tools