The picture for filesystem performance on mechanical/ rotational hard drives is different once again. These are the slowest type of consumer storage devices, and we expected our tests here to highlight any potential design weaknesses of our filesystems. And we were right. During both the Postmark test and the dataset unpacking test, we had Ext4, Btrfs and XFS performing equally well. In Postmark, we saw Reiser5 lagging behind by around 30%, while NTFS looked downright ugly. But when it came to dataset unpacking performance, there was a different result: Reiser5 dropped to its worst figures and performed just as poorly as NTFS. It looks as though the next-gen design crafted by Mr Shishkin is having trouble handling small I/O on mechanical hard drives. Although it looked much better on the SSD drive, this time Reiser5 fell into a pit – it was horribly slow. If you happen to use a rotational HDD, consider choosing another filesystem from our usual suggestions: Ext4, Btrfs or XFS. Any of those three options would do a good job on a mechanical drive thanks to their low fragmentation and efficient block allocation. We tend to trust Sysbench, which indicated that in the long run XFS is a slightly better choice for traditional HDDs. But even a multi-snapshot setup using Btrfs on such a drive is good idea, because that filesystem can autodefrag itself (it’s a mount option). Meanwhile, the only reason to use NTFS on such a drive is compatibility – if you frequently need to move files between a Linux and Windows machine, for example. That’s the main reason why people keep using this lethargic filesystem. Hopefully, things will change when Linux kernel 5.15 starts being adopted by mainstream Linux distros.