Btrfs Seeing Nice Performance Improvements For Linux 5.17

With the Btrfs file-system popularity ticking back up, that seems to be helping upstream enthusiasm and development efforts as with Linux 5.17 there is yet more exciting work.

Btrfs for the Linux 5.17 kernel has prepared another round of performance optimizations on top of tuning found in prior kernel versions. There is also some new core features and never-ending work on code clean-ups and other underlying improvements.

On the performance front for Btrfs with Linux 5.17 there is now less metadata needed for directory logging, which can mean directory deletion is now 20~40% faster.

As another win for this next kernel version, free space tree entries are also now indexed and searched by size which can lead to around a ~30% latency reduction and about 30% lower search times.

For those using Btrfs in zoned mode, the zone information is now cached during mount to speed up repeated queries by about 50%. Separately, there is also less contention now in the tree node locking code when inserting a key and no splits are needed, yielding around a 1~20% improvement with the FS-Mark benchmark.

Btrfs is looking buttery good for 2022.

Outside of the performance work, Btrfs' send code has been adapted to work with concurrent block group relocation, a new "balance paused" mode to allow adding a device to a file-system with paused balance, a new sysfs file for the FSID stored in the per-device directory, extent tree v2 preparations, dropping of the readahead framework, and a variety of other fixes and code improvements.

See this pull request by SUSE's David Sterba for the full list of Btrfs feature patches for the v5.17 merge window.