4 comments

[ 0.22 ms ] story [ 23.8 ms ] thread
From 2020, uses an external 250GB SATA drive over USB, and just tests simple single-file copying. So not terribly relevant if you have a modern internal M.2 NVMe.

That said, it has a very nice overview in terms of ability to extend and shrink partitions with the various file systems.

This is a terrible, terrible article.

- First of all, it's from 2020. Pretty much _every_ filesystem driver/progs has had several performance-related updates since then, so these results are completely irrelevant today.

- ReiserFS is already gone from the kernel, which further shows the age of this article.

- The in-kernel ntfs3 driver is now the preferred NTFS driver. Since the article didn't mention it, the author is likely using the old ntfs-3g (FUSE) driver, especially considering the age of the article. There's a HUGE performance difference between ntfs-3g and ntfs3.

- The tests themselves are completely rubbish:

    - A simple "cp" command is neither accurate nor sufficient; at the very least, a 'sync' should've been run immediately afterwards to flush the primary fs write buffers.   

    - They should've also freed the pagecache, dentries, inodes and disk cache before running each test.

    - A TRIM command should've also been run between tests, since this is an SSD.   

    - The source of data should be a RAM drive so that read speed is accurate and consistent every time.   

    - SSD read operations aren't tested at all.   

    - Random I/O operations aren't tested at all.   

    - Various FS mount options haven't been listed, which can have a significant performance impact (eg compression, atime state etc)   

    - A professional benchmark tool like fio or PTS should've been used.   

TL;DR: This is a completely rubbish and irrelevant article that no one should be reading in 2025.
The random access times are the biggest improvement for SSDs and he didn't mention it.