| Level | Usable | Efficiency | Can lose | Read | Write | Rebuild risk | Cost / TB |
|---|
A degraded array can die twice over. It can hit an unreadable sector while reading every remaining drive end to end — the risk shown above. Or another drive can fail while the first one is still rebuilding, and single parity has nothing left to give. The rebuild window is not short: at half a drive's write speed a 16 TB disk resilvers for the best part of two days, and those are the hardest two days the remaining drives will have all year.
The gap between those two curves is the whole modern RAID argument. Single parity was designed when drives held tens of gigabytes and a rebuild took minutes. The unreadable-sector rate per bit has barely moved since; the number of bits has gone up by orders of magnitude. At the consumer rating of one sector per 1014 bits, eight drives in single parity cross a coin toss at about 1.2 TB a drive and are pinned against the top of the chart long before the sizes you can actually buy — which is why the sector-rate control matters as much as the level does. Double parity does not merely halve that risk; it removes the sector term entirely, because the second parity can reconstruct the bad block, and leaves only the chance of a second mechanical failure inside the window. That is a different order of unlikely, not a smaller number of the same kind.
Two things here are worth distrusting, in opposite directions. The sector model is pessimistic: it treats the manufacturer's rating as a measured rate and every bit as an independent coin toss, and real arrays rebuild successfully far more often than it predicts — read it as "this is no longer a rare event", not as a forecast. The failure model is optimistic: it treats drive failures as independent, and they are not. The drives in an array are usually the same model from the same batch, the same age, at the same temperature, having done the same work — the textbook recipe for correlated failure — and the rebuild is a load spike applied to every survivor at once. The second-failure column is a floor, not an estimate.
Eight 16 TB drives in RAID 5: 112 TB usable, one drive may fail. It does. Rebuilding it at half write speed takes 44 hours — nearly two days degraded.
In that time the array must read 112 TB off the seven survivors, about 9 × 1014 bits. At the consumer rating of one unreadable sector per 1014, that is not a risk so much as an expectation: 99.99%. Switch the sector-rate control to the enterprise 1015 figure and it falls to 59% — still a coin toss you would rather not be offered.
The mechanical risk is far smaller. At 1.5% a year, seven drives over a 44-hour window give about 1 in 1,900 that a second one dies mid-rebuild. Small — but it is the mode that takes the whole array rather than one file.
The same eight drives in RAID 6 spend one more drive on parity, and in exchange the sector term vanishes and two mechanical failures are needed inside the same window: 1 in 8 million. That is the trade, and at these drive sizes it is not close.
Splits data across all drives so reads and writes happen in parallel: maximum speed and capacity. With zero redundancy, a single failure destroys everything. Speed only, never for data you care about alone.
Keeps complete copies on separate drives. Fast, simple and very safe, at the price of half your raw space or less. The performance favourite for databases and VMs.
Recovery information lets a failed drive be rebuilt from the others. Far more space-efficient than mirroring: one or two drives’ worth of parity protects the whole array.
A rebuild reads every surviving drive, and on large drives the chance of an unrecoverable read error mid-rebuild climbs fast. Two-parity levels (RAID 6, Z2) survive it; single parity may not.
It protects against drive failure, not against deletion, ransomware, fire, theft or a controller frying the whole array: those replicate instantly across every drive. You still need real backups, ideally off-site. The 3-2-1 rule is on the Backup Calculator.
After one failure, RAID 5 must read every remaining drive. On big consumer drives the odds of hitting an unrecoverable read error partway through climb past 50%. That is why RAID 6 (or RAID-Z2) is the modern default for large arrays.
Redundancy has to live somewhere. RAID 5 spends one drive on parity, RAID 6 two, mirroring half. That “lost” capacity is exactly what lets the array survive a failure: the price of safety, not waste.
For a handful of small drives, RAID 5 is fine. For large drives (8 TB and up) or many of them, prefer RAID 6 or RAID-Z2; the rebuild-risk readout shows how single parity gets dangerous as arrays grow.
Most parity RAID treats every drive as the smallest one, stranding the extra space on larger drives. This tool assumes matched drives; for mixed sizes, Unraid or ZFS cope more gracefully, though matching your drives is almost always simpler.
No — and this is the most important thing to know. RAID protects against drive failure, but not against deletion, ransomware, fire, theft, or a controller frying the whole array. You still need real backups, ideally off-site.
Redundancy has to live somewhere. RAID 5 spends one drive on parity, RAID 6 two, mirroring half. That "lost" capacity is exactly what lets the array survive a failure — it's the price of safety, not waste.
For a handful of small drives, RAID 5 is fine. For large drives (8 TB+) or many of them, prefer RAID 6 / RAID-Z2 — the rebuild-risk readout shows why single parity gets dangerous as arrays grow.
Most parity RAID treats every drive as if it were the smallest one, so the extra space on larger drives is stranded. This tool assumes matched drives; for mixed sizes, Unraid or ZFS handle it more gracefully. Matching your drives is almost always simpler.
Releases in which this page changed, newest last. Derived from the archived copy of every release, not from notes written afterwards — so it reflects what actually shipped. Site-wide passes are left out; they are in the full changelog.