[cmurf]

One of my long time Btrfs raid1 backups is 99% full. Writes still go at full speed (the much reduced speed you expect to get writing to the interior tracks of spinning disks). But these do avoid that nastiest form of fragmentation due to it only receiving snapshots from another filesystem, making it act like a tape backup until full. Deleting snapshots produces large contiguous holes of space for sequential writes.

This form of snapshotting also scales up well. I've had hundreds of snapshots with no performance reduction, and deleting them is fast also. In cases with many changes in between snapshots, it results in much more complicated metadata. And now while making a snapshot is still cheap and fast, deleting older ones starts to become more expensive for the backref walk.