What are some considerations for selecting datastore size?

[sel57]

I realize there's no one size fits all for datastore sizing these days, but I'd like to know on what criteria I should be basing my opinions.

I'm on ESXi 6.0, and still rocking 2 Tb datastores. For servers requiring larger disks (mostly MS SQL or Oracle db's), I use RDM's as to not bogart an entire datastore or create one bigger than my norm. I had a backup vendor challenge me as to why I was still using RDM's and smaller datastores. I suspect his comments may have partly been because most backup platforms still struggle with RDM's, but I wondered if his comments held any weight and I should be rethinking my architecture decisions.

What criteria do you include/use when choosing how big your datastores should be?

[continuum]

Here is the summary of the lessons I learned doing VMware recovery work:
your datastore is too large when:
- you can not completely evacuate the content to other existing datastores
- you can not completely evacuate the content during friday night / during one weekend
- you have only one
When ever I see a new case and see 2TB datastores I think "good"
Another important lesson I learned:
- immediatly stop using your datastore  if it displays a size larger than 64TB
About RDMs : RDMs are significantly more stable than any VMDK could be as they effectively bypass the VMFS-overhead.
In other words - even with all VMFS-data lost - RDMs are still fully functional.