Linux 4.11 File-System Tests: EXT4, F2FS, XFS & Btrfs

[edoantonioco]

I feel bad for btrfs. It will never be the default file system for desktop usage with those numbers. EXT4 still seems to be the way to go.

[Delgarde]

I am more and more considering XFS as an alternative to EXT4...

What's your use case? What I'm seeing in the benchmarks is that XFS is quicker in some tests, EXT4 in others - so any decision should be based on which set of tests is more like what you need, right?

[wargames]

A file system comparison can never be complete without the "killer" filesystem.

[Ropid]

I feel bad for btrfs. It will never be the default file system for desktop usage with those numbers. EXT4 still seems to be the way to go.

There's just no way it could beat a normal file-system for performance because of its copy-on-write default behavior, but that feature is also exactly why you'd want to choose btrfs over something normal. I guess btrfs could end up as the default if it's needed for something neat, like backups with a comfortable way to browse through them like that Time Machine stuff from Apple.

[dkasak]

I feel bad for btrfs. It will never be the default file system for desktop usage with those numbers. EXT4 still seems to be the way to go.

I've used it, on-and-off, as a desktop / server FS since many years ago. It's not the fastest, to be sure. But it does have some very nice features, eg better recovery of corrupted data in a RAID setup, internal LVM, and it's particularly robust over hard resets. I'm currently using XFS, as Greenplum is only supported on XFS, though I'm thinking of going back on my laptop.

[nuetzel]

Maybe the regression stems from this:
"mm, page_alloc: only use per-cpu allocator for irq-safe requests"
So it was reverted.

https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=d34b0733b452ca3ef1dee36436dab08b8aa6a85c

Please retry with 4.11-rc8...

[Marc Driftmeyer]

The average Linux User doesn't touch any of these products you're benchmarking against their respective filesystems. I suppose you can only work with what's available but benchmarking GNOME, KDE, and their main applications using the following filesystems would have far greater value.

[stiiixy]

I am using XFS for large file storage for video and music and EXT4 for the smaller stuff, as they both handled those respective use cases better for my simple set up. I'm well overdue however for another ring of tests, especially when BTRFS brings out a more stable R5/6.

[Guest]

What's worth to note you benchmarked 4.9 with deadline scheduler and 4.10, 4.11 witch CFQ.

[treba]

Well, the decision for/against a next-gen filesystem should imo not be made on the ground of raw performance. I use BTRFS and ZFS alot these days because their features totally outweigh the performance penalty for me, especially checksumming (I'd never use a big NAS without it anymore), snapshots, compression and deduplication.
Still good to see BTRFS getting more competitive on performance grounds. But performance is barely problem I think.