RAID 10 is a mirror of stripes: your data is split across drives for speed, and every piece exists twice for safety. Here’s how that actually works, which failures it survives, which combination quietly kills it — and the one button that turns a recoverable array into a lost one.
Take four drives. Pair them into two mirrors, then stripe your data across the pairs. Every block lives on exactly two drives — which is both the strength and the fine print.
The stripe half of the design means reads and writes spread across multiple drives at once, so the array is fast — genuinely fast, with none of the parity mathematics that slows RAID 5 and 6 writes down. The mirror half means a failed drive is rebuilt by simply copying its twin, which takes hours rather than the day-plus a parity rebuild can need, and puts far less strain on the surviving drives while it happens. That combination — quick under load, quick to heal — is why databases, virtual machine hosts and busy NAS units so often sit on RAID 10.
The cost is capacity: everything is stored twice, so half your purchased space is spent on the second copy. And there’s a subtler cost worth naming — the safety is positional. The array doesn’t survive ‘any two failures’; it survives failures that land politely, one per mirror.
Two facts account for most of the RAID 10 arrays that reach our bench.
Fact one: mirrors have partners. Lose drive 1 and drive 3 — different pairs — and the array limps on with a complete copy of everything. Lose drive 1 and drive 2 — the same pair — and the blocks that lived on that mirror have no surviving copy anywhere in the array. The odds sound comfortable until you remember paired drives are usually the same model, from the same batch, doing the same work — when one goes, its twin is the most likely drive in the enclosure to follow. Age-matched pairs fail together far more often than probability suggests.
Fact two: rebuilds are not neutral. When a controller offers to rebuild — or a degraded array gets a replacement drive slotted in hot — it starts a marathon copy from a survivor that may itself be developing faults. A rebuild against a weakening twin can push it over the edge mid-copy, converting a one-drive problem into the same-pair disaster above. And if the wrong disk gets pulled during the panic, the controller can begin writing over the exact drive that held the last good copy. The rule when an array is degraded and the data matters: power down, label every drive with its slot number, change nothing. On the bench, each member is imaged read-only first, and the array is reassembled from the images — so no rebuild ever gambles with the originals. That’s the heart of our RAID recovery work, and it applies to RAID 10 in QNAP and Synology boxes just as much as rack servers — see also NAS recovery.
The half-capacity rule lands differently at different scales, and it’s worth doing the arithmetic before buying the enclosure. Four 8 TB drives: 16 TB usable — the classic small-office build, two mirrored pairs, tidy. Six drives: 24 TB usable from 48 bought — and now three pairs share the striping, so performance climbs with the bay count. Eight bays: 32 from 64. Compare the RAID 6 route at the same eight bays — 48 TB usable with any-two-drive protection — and the honest trade emerges: RAID 10 buys speed and fast, gentle rebuilds; RAID 6 buys capacity and tolerance for any two failures rather than the right two. Neither is wrong; paying for one while needing the other is.
Half. Every block is written twice — once to each side of a mirror — so four 4 TB drives give you 8 TB of usable space, not 16. That 50% overhead is the price of keeping a live second copy of everything, and it’s why RAID 10 is chosen for speed and resilience rather than capacity.
They optimise for different things. RAID 5 gives you more usable space from the same drives and survives any single failure; RAID 10 gives faster writes, much faster rebuilds, and can survive multiple failures — if they land in different mirrored pairs. For busy databases and virtual machines, 10 usually wins; for bulk storage on a budget, 5 or 6.
It depends entirely on which two. Lose one drive from each of two different mirrors and the array still has a full copy of everything — recoverable, often still running. Lose both drives of the same mirror and that slice of the data has no surviving copy in the array; recovery then means working directly on the two failed drives themselves, which is exactly the work a lab exists for. Either way: power down and don’t force a rebuild.
It cuts the dangerous window — with one caution. A hot spare lets the rebuild start the moment a member dies, shrinking the hours a pair spends unprotected, and on RAID 10 that rebuild is a gentle single-drive copy rather than an array-wide ordeal. The caution: automatic rebuilds act instantly on whatever the controller believes, so a glitching-but-alive drive can trigger a rebuild against its equally tired twin at the worst moment. Spares suit arrays that are monitored; they don’t replace monitoring.
Free diagnostic on multi-disk arrays at our Belfast lab: every member imaged read-only, the array rebuilt from copies, and a written quote before any work.