QTS proudly reports your mirror is healthy. Underneath, both halves of that mirror may be two files on the same physical disk. This page unpacks the layering honestly, so you know exactly what your RAID is — and is not — protecting you from.
RAID inside a virtual machine obeys the same rule as RAID on real hardware: redundancy only counts when the copies live on genuinely separate failure domains. On a real NAS that means separate disks. In a VM it means separate physical storage underneath the virtual disks — and that is exactly what the tidy settings screen does not show you. Walk the layers below before you trust any pool you build in a lab VM.
👆 The trap in one picture — the mirror is real, but both of its legs stand on the same physical disk (diagram recreated for illustration)
When you create a pool in a virtual QTS, QTS sees disks and treats them exactly as it would in a real NAS. But those disks are files — VMDKs — sitting on the hypervisor's datastore. QTS cannot see through them. It does not know whether its two mirror legs are on two physical drives, two arrays, or the same ageing disk. The pool is honest about the layer it can see, and silent about the layer that actually fails.
Here is the classic build: one datastore, two VMDKs, one RAID 1 pool. It survives exactly one failure — a corrupted VMDK file — and that is a failure that almost never happens on its own. The failures that do happen — the physical drive dies, the datastore corrupts, the host's controller fails — take both legs in the same instant. The mirror was real. The protection was theatre. Nothing in QTS will ever warn you, because from QTS's chair, everything genuinely looks fine.
The same design becomes honest the moment the legs truly separate. Two VMDKs on two datastores backed by different physical disks or arrays really do survive one disk failure. So do passed-through physical disks, one per leg. The rule is mechanical: trace each virtual disk down to the physical device that would have to die to kill it. If both traces end at the same device, you have one copy with extra steps.
Virtual disks add a second, quieter trap. A thin-provisioned VMDK promises the VM a big disk while only using the space the data actually needs — until the datastore fills. When a datastore reaches zero free space, VMware pauses the VM to protect it, and your NAS freezes mid-write. Keep real free space on the datastore, watch it like you watch the pool itself, and remember that QTS's cheerful free-space figure describes the virtual layer only.
Plan capacity twice. Inside QTS: pool size, volume size, snapshot reserve. Underneath it: the datastore must hold those VMDKs plus their growth plus VM snapshots taken before updates. A pool that reports 2 TB free is no comfort when the datastore beneath it has 40 GB left. The offices that get bitten are never the ones that planned badly inside the NAS — they planned only inside the NAS.
So what should you actually do? Match the design to the data. Learning and testing: one datastore, any layout, no guilt — just know the mirror is decorative. Convenience copies of things that exist elsewhere: acceptable, with a backup outside the host. The only copy of anything that matters: not here, at any RAID level. That data belongs on real hardware — or at minimum on separated physical storage with an external backup — and the backup page in this series shows what that discipline looks like.
Watch the video for the storage setup shown live — then apply the layering test from this page to your own build.
▶️ Watch Full VideoBecause health is reported at the virtual layer. If both VMDKs sit on one datastore on one physical drive, the drive's failure kills both legs at once. The mirror only protects against a failure of one virtual disk file, which is the rarest failure of all. Trace each disk to its physical device — that is the real answer.
Put the two legs on genuinely separate physical storage: different datastores backed by different disks or arrays, or passed-through physical disks. Then one physical failure really does leave the other leg standing, and the RAID level in QTS finally means what it says.
It adds a second one. A thin VMDK grows as data arrives, and if the datastore fills completely, VMware pauses the VM to protect it — your NAS freezes without anything being broken. Keep real free space under the virtual disks and monitor the datastore, not just the pool.