Thin provisioning is one of those ideas that seems really nice, but has some caveats that can really make it tough to use. First, volumes that are thin provisioned rarely shrink again. There are some ways to do it, especially with deduplication, but they’re I/O and labor intensive. Over time your volume will always get bigger, and rarely gets smaller. Second, thin provisioning introduces variability in your storage, especially in environments that allow customers to change, add, and delete data. It’s very possible to be caught short.
Of course there are upsides, too. In the virtual world it means a VM can look larger than it is, but save space that isn’t in use. That saves admin time extending volumes. There are also some newer technologies, such as the VMware VAAI “UNMAP” command that are starting to work to re-thin volumes in an automated fashion.
So are you using thin provisioning? Where are you doing it: on the array, on the hosts, or both? How are you managing the risks associated with it?