Btrfs In Linux 6.5 May Bring A Cumulative Performance Improvement

[KyleSanderson]

ZFS and BTRFS are both terrible for performance. Don't get tricked, there's plenty of liars everywhere. XFS is the most reliable, with ext4 still edging out for the single thread usecase.

[Developer12]

Wake me up when they fix RAID5/6.

At this rate bcachefs will get merged first and it's erasure coding and checksums will make btrfs totally irrelevant to anyone not already running ZFS instead.

[Danny3]

They or somebody else should also upgrade the Zstd code from v1.5.2 (updated in Linux 6.2) to v1.5.5, upstream, which has good performance improvements:

https://github.com/facebook/zstd/releases/tag/v1.5.5

[some_canuck]

Wake me up when they merge and switch to the new on disk format.

Wake me up when they fix RAID5/6.

[pkese]

So when should I switch from ext4 for my root partition, is the file system ready yet and is it SSD friendly in terms of the amount of writes compared to ext4?

It's fine and it's SSD friendly.
I've been using it for about 10 years on all my machines including several production servers and have never had any problems whatsoever.

You do trade a bit of performance for some extra features, like checksums, snapshots, subvolumes, reflinks, etc, but I think it is worth it - disk I/O performance is usually not an issue lately.

I wouldn't use it on unstable or highly overclocked machines, because it can be quite unforgiving with regards to bit flips due to CPU or memory instability.

[rob-tech]

So when should I switch from ext4 for my root partition, is the file system ready yet and is it SSD friendly in terms of the amount of writes compared to ext4?

[cynic]

check user visible changes here:

https://btrfs.readthedocs.io/en/latest/Kernel-by-version.html#dec-2022

i think this is the page you were thinking of. This block group tree was split out from the larger extent v2 work, since it was ready to go, and gives much needed improvements. I'm really not sure what the state of this bigger effort is.

https://josefbacik.github.io/kernel/...bal-roots.html

thank you!

the page I remebered was another one. Maybe was on github?
anyway these two links are very useful!

didn't know they already merged a lot.

[fitzie]

The new on disk format is a must to get things like quotas+snapshots usable.

my mount time went from 5 minutes to 5 seconds.. prior to that I had to keep increasing systemd timeout settings (DefaultTimeoutStartSec) to keep up with how bad it was.

[fitzie]

I remember there was a working progress page for this, but cannot find it anymore.
do you have any link?

check user visible changes here:

https://btrfs.readthedocs.io/en/latest/Kernel-by-version.html#dec-2022

i think this is the page you were thinking of. This block group tree was split out from the larger extent v2 work, since it was ready to go, and gives much needed improvements. I'm really not sure what the state of this bigger effort is.

https://josefbacik.github.io/kernel/btrfs/extent-tree-v2/2021/11/10/btrfs-global-roots.html

[darkbasic]

The new on disk format is a must to get things like quotas+snapshots usable.