
In short
- RAID protects against disk failure, not against deletion, ransomware or a failed rebuild.
- RAID 6 or 10 is safer than RAID 5 for large disks.
- A failed array is rebuilt from disk clones, never on the original disks.
The building blocks
- Striping: data is split into blocks (stripes) spread across several disks. Fast, but no protection on its own.
- Mirroring: every block is written to two disks. Simple and robust, but you lose half the capacity.
- Parity: an extra block, calculated from the data blocks (an XOR for RAID 5, plus a Reed-Solomon calculation for RAID 6), lets missing blocks be recalculated.
The levels
| Level | Min. disks | Usable capacity | Survives |
|---|---|---|---|
| RAID 0 | 2 | 100% | No failure at all |
| RAID 1 | 2 | 50% | One disk |
| RAID 5 | 3 | (n − 1) / n | One disk |
| RAID 6 | 4 | (n − 2) / n | Two disks |
| RAID 10 | 4 | 50% | One disk per mirror |
RAID 0
Pure striping. Every file of more than one stripe has pieces on every disk. One failed disk takes the whole volume down. See RAID 0 (striping) data recovery.
RAID 1
A mirror. Both disks hold the same data. Easy to recover, but a mirror copies deletions and corruption instantly. See RAID 1 (mirror) data recovery.
RAID 5
Striping with one rotating parity block. The most common level in small NAS units and servers, and the most fragile during a rebuild. See RAID 5 data recovery and RAID 5 rebuild failed: what happened and what to do now.
RAID 6
Two parities, P and Q. It survives two failures, which makes it the sensible choice for arrays of large disks. See RAID 6 data recovery.
RAID 10
Mirrors striped together. Fast for databases and virtual machines, and quick to rebuild, but two failures in the same pair are fatal. See RAID 10 data recovery.
Hardware RAID, software RAID and NAS
- Hardware controllers (Dell PERC, HPE Smart Array, Broadcom MegaRAID, Adaptec) store the configuration on the disks and in the controller. A dead controller or a lost configuration can hide perfectly healthy disks.
- Software RAID (Linux mdadm, Windows Storage Spaces, ZFS, Btrfs) stores everything on the disks themselves. See Reassembling a Linux mdadm RAID without making it worse.
- NAS units are software RAID with their own layers on top: SHR on Synology, LVM and thin volumes on QNAP. See The complete guide to NAS: how they fail and how to protect them.
How arrays fail
- A second disk fails during the rebuild after the first.
- A disk dropped out weeks ago and nobody noticed the alert.
- The controller loses its configuration after a firmware update or a replacement.
- Someone pulls the wrong disk, or recreates the array "with the same settings".
- A power cut during writes leaves parity inconsistent (the write hole).
What to do when an array fails
- Stop writing to it. Do not rebuild, recreate or initialise anything.
- Note the slot of every disk and the controller or NAS model.
- Export the logs if you can do so without changing the array.
- Have the disks cloned. The array is then rebuilt virtually from the clones: disk order, stripe size, parity rotation, offset. See RAID data recovery.
RAID is not a backup
RAID keeps a system running when a disk fails. It does nothing against deletion, overwriting, ransomware, theft, fire or a controller that writes garbage to every disk at once. Keep at least one copy elsewhere, ideally offline.
Frequently asked questions
Which RAID level should I choose for a four-bay NAS?
RAID 5 gives more space, RAID 6 or RAID 10 more safety. With large disks, RAID 6 is the safer choice, and you still need a backup.
Can a failed RAID always be recovered?
Not always, but very often. What matters most is not making changes to the disks before they are cloned.
Is your data important?
Do not try anything that could make the fault worse. The diagnosis is free and you get a fixed quote before any work starts.



