Ideas for supporting shingled magnetic recording (SMR)

> Right, That's exactly what most people fail to recognise with ZFS - the management interface combined filesystem and device management, but the internal implementation was still very strongly layered in the traditional namespace/block/device layers.

> [...] and that's a mistake we made 6-7 years ago by taking the btrfs path.

BTRFS actually has a strong block device layer internally, it just doesn't provide the same interface as the classic Linux block device layer, just like ZFS has a strong block device layer internally, which isn't the same as the classic Solaris block device layer.

The reason for this is quite simple: the classic block device interfaces does not provide everything BTRFS and ZFS need. While it certainly would have been possible to rewrite the old block device code to provide the extended functionality needed, you would have a hard time convincing enterprise users to trust their data to the new and unproven block device code, when the old and proven version can is perfectly capable to host their ext3 file system...