We've been doing some testing with vCloud Director 5.1.1, with vSphere 5.1 and Storage DRS.
We have a Datastore Cluster consisting of 3 x 500GB NFS volumes, making 1.5TB in the cluster.
We have Fast Provisioning enabled but not Thin Provisioning for the test OrgVDC. The OrgVDC is given 800GB of storage
As you'd expect, vCD won't let a user create a single drive with more than 800GB (because that's the max allowed). It also won't allow a disk of more than 500GB (larger than a single datastore). That's fair enough, but the error message isn't informative.
The Org admin is able to create a single VM with 2 x 300GB drives (totalling 600GB). This doesn't throw an error, and creates the VMDKs on one NFS datastore. NFS datastores are thin provisioned by default. The datastore shows as 120% provisioned but much less used.
When the VM writes enough data to those two drives, the VMDKs fill up, and eventually the datastore hits 100% usage and all the VMs on that datastore pause.This is a "Bad Thing".
Storage DRS moves other things off the datastore, but can't move the disks of the single VM to different datastores unless we manually change VM-specific settings in vCenter which is rather defeating the point of having vCD.
One workaround would be to never have more than one NFS datastore per Datastore Cluster, but that seems rather limiting as we'd like to spread storage over multiple devices.
If I enabled "Thin Provisioning" at the OrgVDC, would that let vCD make better decisions about over-allocating storage?
Is there a Best Practice guide for datastore sizing? Do individual datastores need to always be larger than the maximum storage any single OrgVDC can take?
Any other suggestions for avoiding this problem?