I really hope it would be a working one, not like xfs where your files may just disappear with no trace (never on Irix, never on any other fs) or like btrfs which may just suddenly go read only and be dead on reboot with no fsck and all data unreachable.
How hard is it to get the basics right? Doesn’t matter how much rice there is if it keeps blowing up.
Me too. I’ve run 30 years with ext and bsd filesystems with no failure. Many years with various UNIX native fs as well. But Linux xfs, reiserfs, btrfs all have resulted in catastrophic failure within a year on several machines. They’re permanently off my list, but I have some hope that someone will get a new fs right.
Features include caching,[4] full file-system encryption using the ChaCha20 and Poly1305 algorithms,[5] native compression[4] via LZ4, gzip[6] and Zstandard,[7] snapshots,[4] CRC-32C and 64-bit checksumming.[3] It can span block devices, including in RAID configurations.
The main takeaway from the article is that the developer’s name is Kent Overstreet, who beat his bitter rival Surrey Underpath, who are both canonically related to famed developer Cornwall Midroad.
As someone else said, it’s similar to btrfs. bcachefs has a lot of functional overlap with btrfs, which is great. There have also been a few benchmarks showing that bcachesfs is faster for some situations (cold-cache warming, IIRC). One of the big advantages over btrfs is that bcachefs’s RAID is more robust - several of btrfs’s RAID levels have been marked as experimental and prone to data loss, for years. There’s been improvement in btrfs RAID lately; the skeptic in me believes this is directly a result of pressure from bcachefs, which is in a position to become a favored fs in Linux.
A lot of the time it obviously takes a little while for userland tools to catch up and for distros to include both the new kernel and userland tools for it into their latest versions but once that is done average users certainly do notice differences. Literally all the features that are talked about a lot like BPF or io_uring or all the features that make containers possible were introduced in a kernel release at some point.
Add comment