Virtualisation hammers storage with mixed, random I/O from many VMs at once — so the best RAID prioritises IOPS and low latency. Here's how to choose for VMware, Proxmox or Hyper-V, and size it in the RAID calculator.
Why virtualisation is IOPS-bound
A virtualisation host runs many VMs whose I/O blends into a heavily random, mixed read/write stream — the 'I/O blender' effect. That makes IOPS and latency, not raw capacity, the limiting factor. The write penalty matters a lot here: parity RAID's ×4–×6 tax throttles the write IOPS many VMs depend on.
So the usual answer is flash plus a low-penalty layout. RAID 10 on SSDs gives the best write IOPS and fast rebuilds; all-flash RAID 6 is viable where capacity matters and writes are moderate, because SSD IOPS are abundant.
Layout by platform
VMware/Hyper-V on a hardware RAID host: RAID 10 on SAS/NVMe SSD for performance, or RAID 6 all-flash for capacity-led estates. Proxmox or TrueNAS with ZFS: use mirror vdevs (striped mirrors) for VM datastores — a RAIDZ vdev only gives one drive's random IOPS, which throttles many VMs, whereas mirrors scale IOPS with vdev count. Keep volblocksize sensible (small RAIDZ volblocksize wastes capacity — see ZFS tuning).
For shared storage across a cluster, this maps to all-flash arrays (NetApp, Pure, Dell PowerStore) — our storage finder sizes those.
View the data behind this chart
| RAID 10 | RAID 6 | RAIDZ2 | |
|---|---|---|---|
| Write penalty | ×2 | ×6 | CoW |
| Random IOPS | High | Medium | ≈ 1 drive / vdev |
| VM datastore? | Best | Capacity | Use mirrors |
Sizing it
Estimate IOPS per VM, multiply by VM count, add headroom, and check the front-end write IOPS the layout delivers in the calculator (it applies the write penalty for you). If RAID 6's ×6 penalty starves writes, switch the calculator to RAID 10 and compare.
Remember the host still needs a backup and ideally array snapshots/replication for the VMs — RAID alone isn't VM protection.