vCenter

We've got an SDRS cluster that is around 80TB, made up of 10 datastores - utilization is 48TB, with 31TB free.

Our largest stretch of free space within the cluster is ~6TB

We have a VM that is around 8TB (total of all its VMDKs) that we want to relocate into this SDRS cluster - when we try to move it we get an error about insufficient disk space, since our largest free space block is less than the size of the VM.

We just recently moved to 6.5 and I swear in the past (back in 5.5 days) when we moved VMs larger than our largest free space SDRS would move OTHER VMs around and create more space (so long as there was space in the cluster for this).  Since upgrading to 6.5 it seems like this no longer occurs - if we don't have a large enough stretch of free space we're just SOL, or am I missing something? We have datastores that are 10TB, so we should have plenty of space for this Vm to move into the SDRS cluster if other VMs would move to other datastores.

I can of course resolve this manually, but I'm really wondering why SDRS isn't sorting it out like it used to.

Hi,

The FAQ below indicates the default space threshold per datastore is 80%. My guess would be that it is not intervening as it would cause >80% utilization of any the datastores, >8/10.

Do you have a 7TB VM you could test that theory with. I guess another option is to tweak the default threshold per datastore?

Considering the size of that beast VM, might be better and faster to do it manually but I do understand why you want it to just work (as it did before). Where the VMs on the DS cluster so large in the 5.5 days?

No P.RDMs attached?

Failing that I'd say it is a bug or needs a better error to say why it is not taking action.

Storage DRS FAQ

http://kb.vmware.com/kb/2149938

Storage DRS generates only one datastore as initial placement recommendation in clusters with physical RDM

http://kb.vmware.com/kb/2148523

sorry for the late reply - I suspect your 80% comment is correct. VM has no P.RDMs, all VMDK based. Unfortunately this is hard to test without creating a bunch of insanely huge VMs and datastores. It's definitely not a normal situation.

To get around this 'problem' I ended up evacuating a datastore, then pulling that datastore out of SDRS. Then relocated the vm in question, and added it back to SDRS. Ultimately there is no real reason for this VM to be managed by SDRS, but we're trying to do some datastore consolation and make sure SDRS is monitoring io/space. More for consistency than anything else.