Bcachefs Lands Another Round Of Fixes For Linux 6.7
On Friday another round of fixes were merged for the Bcachefs file-system for the in-development Linux 6.7 kernel.
This week's round of Bcachefs fixes isn't as big as their second round of updates merged for Linux 6.7 but still does contain some notable fixes. The new Bcachefs fixes include taking care of the builds being broken when Control Flow Integrity (CFI) is enabled, erasure coding for the file-system is now marked as an "extra-experimental" feature with some incompatible disk space accounting changes expected, several fixes for durability-related issues, and several other fixes to this file-system driver.
With the erasure coding being treated as ultra-experimental and known breakage expected ahead for the disk format, the "erasure_code" feature is now hidden behind the new BCACHEFS_ERASURE_CODING Kconfig option.
The full list of Bcachefs file-system fixes merged this week can be found via this merge.
In case you missed it on Thursday was my latest round of Bcachefs benchmarks. Bcachefs contributor Daniel Hill also emailed in some feedback and details on the latest testing:
It's great to see the progress being made on Bcachefs and the developers being aware of areas for (performance) improvements moving ahead.
This week's round of Bcachefs fixes isn't as big as their second round of updates merged for Linux 6.7 but still does contain some notable fixes. The new Bcachefs fixes include taking care of the builds being broken when Control Flow Integrity (CFI) is enabled, erasure coding for the file-system is now marked as an "extra-experimental" feature with some incompatible disk space accounting changes expected, several fixes for durability-related issues, and several other fixes to this file-system driver.
With the erasure coding being treated as ultra-experimental and known breakage expected ahead for the disk format, the "erasure_code" feature is now hidden behind the new BCACHEFS_ERASURE_CODING Kconfig option.
The full list of Bcachefs file-system fixes merged this week can be found via this merge.
In case you missed it on Thursday was my latest round of Bcachefs benchmarks. Bcachefs contributor Daniel Hill also emailed in some feedback and details on the latest testing:
"I'm doing a bit of work on bcachefs with Kent, he informs me that we only do full extent checksums and we need to work on per sector checksums, this is why you're seeing degraded performance in some of those tests, There's something I've been seeing, and I'm not quite sure why, but crc32c never seems to use hardware acceleration if you compile bcachefs as a module.
Hardware crc32c is needed to keep up with enterprise SSDs, especially single threaded loads, and per-extent checksums is still rather
excessive for every read too, hopefully we can fix that in a new version."
It's great to see the progress being made on Bcachefs and the developers being aware of areas for (performance) improvements moving ahead.
24 Comments