By manufacturer
Dell PowerEdge HPE ProLiant Lenovo ThinkSystem Fujitsu Primergy Supermicro IBM System x / Power Acer AltosRAID 6: dual parity. Recovery beyond the two tolerated disks.
RAID 6 survives the failure of two disks thanks to dual parity (P + Q). When three fail, the array is marked failed, but recovery is still possible by working on the surviving disks and rebuilding parity in the correct order.
P + Q, computed with Reed-Solomon.
RAID 6 keeps two parity blocks for every stripe: the P block (plain XOR, the same as RAID 5) and the Q block (computed with Galois field coefficients, Reed-Solomon formula). This makes it possible to rebuild up to two missing disks.
On a triple failure, the surviving disks still hold information that is useful for recovery, even though they are not enough to rebuild the data directly — especially if the third failed disk can be recovered in the lab.
Disk recovery + algebraic rebuild.
- Physical recovery of the failed disks (PC-3000, head transplant in a clean room where required).
- Parameter identification: slot order, chunk size, controller parity algorithm (it varies across Dell PERC, HPE Smart Array, LSI MegaRAID, Adaptec).
- Algebraic rebuild on the clones of the available disks.
- Filesystem recovery on the rebuilt volume.
The questions we are asked most often.
How common is a triple RAID 6 failure in production?
Rare, but not exceptional. Typical causes: batches of disks from the same source failing at around the same time, an abnormal electrical event, a long rebuild during which a second or third disk gives out under the additional workload stress of the rebuild itself.
RAID 6 or RAID 5 + hot-spare?
It depends on the workload. RAID 6 has a higher parity overhead (double parity write) but greater resilience. RAID 5 + hot-spare offers better write performance, but if another disk fails during the rebuild, the array is lost. On large arrays with high-capacity disks, where a rebuild can take days, RAID 6 is almost always the rational choice.