CrystalDisk Mark works, in total, eight performance tests to measure how hard our hard drive or SSD works, four read and write four (so there are four stages). Each of them gives us different performance data, do you know what is measured for each one and how to interpret it?
Understanding the concepts of tests
This is a screenshot of any tool bench, this time the PCIe 3.0 x4 NVMe SSD.
In fact, from the four categories used by CrystalDisk Mark to measure performance, we have to distinguish two approaches to them: linear and random.
- Sequence (SEQ): data is read or written in chronological order, thus avoiding looking at the device storage component where it is required to read or write. This type of test represents when we copy something to a disk, or read a file that is too large for it.
- Random (RND): the information is read or written to random parts of the device, following the "free and clean first" rule. So in this test, the actual performance is measured when we have to handle multiple small files.
Within these categories, you must specify additional parameters when it comes to what each test does:
- Line Depth: This data represents a buffer type or read or write cache, and tells us the number of execution commands that can be performed per cycle. The higher the number, the better the performance.
- Sticks: is the number of reading or writing tasks that work every minute in parallel, that is, at the same time. It works much like a string of processors depending on the concept, although in terms of empirical it is very different because if we force a disk to produce multiple threads at once, it can be easily fulfilled.
CrystalDisk Mark Testing
Now that we're clear about the concepts of the test, let's look at what each test means and how we should interpret the results.
- SEQ 1M Q8 T1: Respectively, for 1 MB files in size, x88 depth line and processing thread. This scenario is probably the right situation, with large files written or read in succession, and will basically show us the highest performance of the app.
- SEQ 1M Q1 T1: In a row, for 1 MB files in size, x1 depth line and thread to process. This is a general situation, and one that will show the performance of the app in intermediate conditions, that is, its daily use.
- RND 4K Q32 T16: Random, with 4KB files in size, x32 tail depth and 16 threads. This test shows us how the device was scanned with small, but multiple files at once and with multiple threads, giving a device edge. Ideal for viewing SSD or hard drive performance in line instances with multiple small files.
- RND 4K Q1 T1: Random, with 4 KB files in size, depth x1 and 1 string. It's a terrible case for a storage device, because as you know based on the size of the file allocation (provided when we format it) as a rule it "doesn't work" for smaller files, and the more we line and the number of threads at the same time, the less. It is natural for this parliament to have a much lower impact compared to the previous one, but the high speed here represents a very effective backup tool for managing small files.
Of course, if we want to measure the performance of a hard disk or SSD under certain circumstances, we can change exams in the settings menu -> Queue and Rows (translation is very bad, but that's what it says).
So we can measure the performance of the benchmarking tool we want, or the tests that come automatically and that we described above are designed correctly to bring the instrument to its limit, but also to show its performance in real-world situations.