Say I have a 30 GiB disk and I'm only using 20 GiB for PVs.

On the cloud, we are able to match the disk-id from the cloud bill to the node, and therefore by combining cloud data with CMMO data, we know there's 10 GiB of unallocated storage that should be attributed to the node (and therefore the cluster, and by means for  COST-3165 and siblings, to OCP projects).

On premise, on the other hand, we don't even know the original disk was 30 GiB because there's no cloud bill to read from, so we only have CMMO data, which only knows about the 20 GiB that were allocated to PVs.

We should think of some way to capture those "missing" 10 GiB in the on-prem case:

• Do we need an agent at the hypervisor level? What to do on bare metal, then?
• Should we introduce some kind of "on-premise node cost model"? It will certainly be a need for Cost Management for RHEL in the on-premise case
• Something else?