The advantage is that you can get all the features you want, even though they aren't all available in one filesystem. In my case, I wanted a reliable filesyste

The advantage is that you can get all the features you want, even though they aren't all available in one filesystem. In my case, I wanted a reliable filesystem, RAID1/mirror support, block-level checksumming, and full-disk encryption. No filesystem provides these on linux right now. My solution was therefore to use ZFS to provide a mirrored, checksummed, reliable volume -- onto which I put a standard LUKS-encrypted ext4 filesystem. In the past I had tried the opposite (LUKS on the bare drives, then a ZFS mirror of the 2 decrypted volumes), but it was kinda annoying to manage, and I don't really need the other ZFS features (like snapshotting). In the grandparent post, the requirement was the ability to expand the RAID volume without rebuilding (something that ZFS doesn't offer), plus checksumming and reliability (which ZFS does offer). So one option would be to use mdadm to manage the RAID array, and then put ZFS on the resultant volume in order to get checksumming. The disadvantages are: extra complexity; extra overhead; more potential points of failure; more management hassle; etc. As soon as encryption for ZFSonlinux is stable, I'll be very happy to drop this filesystem stacking in favor of that!