By manufacturer
Dell PowerEdge HPE ProLiant Lenovo ThinkSystem Fujitsu Primergy Supermicro IBM System x / Power Acer AltosRAID 10: striping of mirrors. Fast recovery as long as the failure does not hit both sides of a mirror.
RAID 10 (1+0) stripes data across multiple mirrors: it tolerates many failures as long as they do not hit both disks of the same mirror set. When that happens — a double failure within the same mirror — the array is failed, but recovery is still possible by working from the healthy side of the surviving mirrors.
Mirror sets + stripe.
RAID 10 = N mirrored disk pairs, striped across. Example with 4 disks: D1 mirrors D2 (mirror set A), D3 mirrors D4 (mirror set B). Data is striped across A and B.
Tolerance: you can lose D1 and D3 (one disk per mirror set) and the array keeps running. You can lose D1 and D2 (both sides of the same mirror) and the array fails.
Recovery from the healthy side + stripe reconstruction.
- Identifying the mirror set at fault: from controller metadata, from disk serials, from the documented physical order.
- Physical recovery of at least one of the two disks in the failed mirror set, where feasible.
- Stripe reconstruction: with the mirror set restored (even partially), recovery of the overall stripe.
A single disk fault does not stop the stripe.
If the failure hits only one side of a mirror, the array keeps running in degraded but operational mode. Replacing the failed disk and rebuilding the mirror is a linear operation, with no destructive recovery. That is why RAID 10 typically recovers faster than RAID 5/6 — as long as the failure pattern stays favourable.
The questions we get most often.
Is RAID 10 worth it over RAID 5/6 in terms of performance?
RAID 10 typically delivers higher write performance (no parity calculation) but costs space: 50% of raw capacity. On write-intensive workloads (databases, virtualisation) it is often the right call. RAID 5/6 has lower overhead and more usable capacity, but a higher write penalty. The choice depends on the read/write mix.
Double failure across different mirror sets: is the data safe?
Yes, RAID 10 theoretically tolerates N-1 failures (on an array of N mirrors) as long as at least one disk survives on each mirror set. The real point of failure is a double failure within the same mirror set.