Every RAID level is a bargain struck between three things you want — capacity, speed and survival — and you cannot have all three. Understanding which one your array traded away explains, precisely, the way it is going to fail. Here is each level in plain English, and the specific death that arrives with it.
A RAID array exists to keep a service running when a disk dies — not to keep your data safe. It cheerfully replicates every deletion, every ransomware encryption and every accidental overwrite across all its members at machine speed. The redundancy is uptime insurance, and the difference between uptime and safety is the single most expensive misunderstanding in storage.
The mechanism worth knowing before the levels make sense is parity. Parity is not a copy of your data; it is a mathematical summary that lets the array reconstruct the missing piece if one drive drops out. It is elegant, and it is also the source of the most common disaster in this entire field — because reconstruction has to read every remaining disk, completely, at the worst possible moment.
RAID 0 (striping). Data is cut into stripes and spread across every disk. Fast, uses every gigabyte you paid for, and has zero redundancy: one member dies and the entire set is gone, because half of every file was living on it. RAID 0 is not a compromise on safety, it is the absence of safety — and yet it ships as the default on plenty of two-bay boxes sold as “backup drives”.
RAID 1 (mirroring). Two disks holding identical contents. Simple, survives one drive dying, halves your capacity. Its quiet flaw is that most people never look at it again: the mirror can sit degraded for months, running happily on one disk, and nobody notices until the survivor fails too.
RAID 5 (striping with single parity). The workhorse: capacity of all disks but one, survives exactly one failure. Its bargain is the harshest of the lot, and the reason is in the next section.
RAID 6 (dual parity). Like RAID 5 but survives two simultaneous failures, at the cost of another disk’s worth of capacity. On the large drives sold today it is the sane choice, and its enemy is complacency rather than mathematics.
RAID 10 (a mirror of stripes). Pairs of mirrored disks, striped together: the speed of RAID 0 with the redundancy of RAID 1, at half your raw capacity. It survives one disk from each mirrored pair failing — and dies instantly if both halves of a single pair go. Which is a real risk when four identical drives from one production batch have been spinning in the same warm cabinet since the same afternoon.
The RAID 5 rebuild. A disk fails, you slot in a replacement, and the array reads every sector of every surviving disk to reconstruct the missing one. On multi-terabyte drives that read is enormous — and drives that have been idling for years in the same enclosure develop unreadable sectors quietly. The rebuild hits one, a second disk drops out, and the array collapses during the repair. A Paisley practice lost exactly this way, and it is the single most common array disaster there is.
The wrong drive. Under pressure, with the amber lights blinking, the healthy disk gets pulled instead of the failed one — and the rebuild starts writing over good data with reconstructed nonsense. We have rebuilt the aftermath: a rebuild started onto the wrong member, with the virtual machines still inside.
The power cut and the firmware update. RAID 0 in a NAS, mid-write, when the lights go out: metadata half-written, array unmountable, and no parity to reason from — a ReadyNAS did precisely this. The controller-amnesia variant is just as common: a card is replaced, or a firmware update rewrites the configuration, and the array comes back claiming its members are foreign.
Deletion, which redundancy cannot save you from. A volume deleted on a healthy RAID 5 is gone from every disk simultaneously — as one Dell PowerEdge found. No parity in the world reconstructs a decision.
The instinct when an array fails is to make it healthy again — rebuild, resync, re-initialise, “repair volume” in the NAS menu. Every one of those actions writes to the disks holding your only copy, and a failed rebuild is far harder to undo than a failed drive. The professional sequence inverts it: image every member disk individually first, reconstruct the array virtually from those images, extract the data, and only then argue about the hardware.
What that needs from you is order. Power the array down. Do not run the rebuild again. Label the drives with the bay they came from — stripe order is everything, and an unlabelled pile of four identical disks is a puzzle we can usually solve but would rather not have to. A NAS can be posted as a whole unit or as its bay-labelled drives; a server’s drives travel out of the chassis, labelled. Array recovery starts from £500 + VAT, the diagnostic is free, and the figure is fixed in writing before anything is touched.
A degraded array is recoverable; a rebuilt-over-the-top array is a much longer story. Stop the resync, label the drives by bay, and let the free assessment map what is really left — RAID, NAS and server work from £500 + VAT, fixed in writing.