A full blown filesystem is overkill for an object store. You could use something like libzbc ( https://github.com/hgst/libzbc https://github.com/hgst/libzbc ) t

A full blown filesystem is overkill for an object store. You could use something like libzbc ( https://github.com/hgst/libzbc https://github.com/hgst/libzbc ) to write directly to the SMR drives on the block level. I believe Ceph now has abstracted the drives away through BlueStore, which simply puts a large RocksDB database on the drive, bypassing most of the functionality a filesystem offers. It should be much easier to make an SMR compatible version of the LSM-tree backend of RocksDB, than writing a full-blown file system.