LVM can use any block level device (anything inside /dev/ like HDD, SSD, NVMes, etc). I will add some quick info about how to create these on the post later today.

But LVM snapshots are per LV basis, so because usually you have a LV (a) per FS, when you snapshot that LV, what is happening, is you are creating another LV (b) in the same VG where all the modified blocks of the original LV (a) are being copied into the snapshot LV (b).

So if you need to separate at MongoDB level, you might want to have different filesystems/LVs for each of the collections you have, and then either also separate using multiple mongo instances (I would find it easier), or specific different folders for the collection so that you can then use them a separate filesystems and be able to use different LV snapshots for each of them.

The great part about having everything on one place is simplicity, but space use might be a problem if you don't want to snapshot everything, yep. Hence separating per instance/collection would be necessary.

I will update more by the end of my day.