Bcachefs File-System Re-Submitted For Linux 6.6

[royce]

BFS is the name y'all looking for

[kreijack]

it's raid5/6 implementation while not yet ready yet won't have design-phase problems like btrfs.

Strong sentence for something that doesn't exist (yet)...

From the Btrfs maintainer:

so regarding the write hole the raid-stripe-tree will address that, though its primary purpose is to enable RAID profiles and zoned devices.

The raid stripe tree is over killing to solve the raid hole.

The straightforward way to solve the raid hole in BTRFS is to have a journal for the writing. There were already some effort in this direction by Liu Bo and Qu Wenruo. However no conclusion was reached.

[1] https://lore.kernel.org/linux-btrfs/[email protected]/
[2] https://lore.kernel.org/linux-btrfs/[email protected]/

My impression is that there is not enough interest to solve this issue to pay a developer to implement the journal write. However Qu is doing some work to cleanup the BTRFS5/6 code.

I don't think that BTRFS has an insolvable RAID5/6 design problem. The problem is only a poor implementation which was developed as simple extension of the other profiles, when instead it is a completely different beast.

[kreijack]

BFS is the name y'all looking for

In the code I saw the bch prefix. BCHFS ?

[EphemeralEft]

there is no comparable solution.
afaik no free linux filesystem offers tiered storage.
it's raid5/6 implementation while not yet ready yet won't have design-phase problems like btrfs.

Linux doesn't have an all-in-one solution like ZFS, but it has all the same technologies via separate block layers / "devices". These layers can be combined in whatever way and order you want to use multiple technologies together. This way, you can use whatever implementation you want for a particular technology. The other nice thing is that anything which you can put on a block device can get those features, without each subsystem/filesystem independently implementing them. So EXT4 actually supports tiered storage, RAID, and encryption. So does XFS and FAT32.

• RAID => DM-RAID, MDADM
• Tiered storage => DM-Cache, Bcache
• Volume management => LVM
  • CoW snapshots
  • Thin provisioning
  • Compression
  • Deduplication
• Encryption => DM-Crypt
• Integrity => DM-Integrity

The only ones that really suck (right now) are DM-Integrity and CoW snapshots via LVM. Compression and deduplication are OK, but not great. Luckily all of those things are done very well by BTRFS. You need 2 or 3 userspace tools to manage all of them—which is admittedly not as sexy—but it's much more functional.

[M@yeulC]

there is no comparable solution.
afaik no free linux filesystem offers tiered storage.
it's raid5/6 implementation while not yet ready yet won't have design-phase problems like btrfs.

Do you really need raid5/6 when you have forward error correction? Especially if the fec data can be smartly distributed on other drives... (Well, you may argue that this becomes raid5/6 again at some point).

Fec is much welcome, it's annoying to be able to detect single bit errors thanks to checksumming but not being able to correct them.

[Barley9432]

[QUOTE=EphemeralEft;n1407400]

Linux doesn't have an all-in-one solution like ZFS,

What? OpenZFS supports Linux.

[Barley9432]

Linux doesn't have an all-in-one solution like ZFS,

What? OpenZFS supports Linux.

[WereCatf]

What? OpenZFS supports Linux.

I believe OP's point was that OpenZFS is an out-of-tree module; it's not in the kernel sources.

[Vaporeon]

bcachefs is a cool name, do not change it.

[stiiixy]

...which is admittedly not as sexy...

Where does one find a SexyFS.