How Long Does a RAID Rebuild Take?

The write is the easy part

Rebuild time is the replacement drive's capacity divided by the rebuild speed. An 8 TB drive at 100 MB/s takes about 22 hours.

That's the optimistic figure — an array doing nothing else. A NAS still serving files often manages a third of that speed, because the same disks are doing two jobs and many controllers deliberately throttle rebuilds to keep the array responsive.

The read is five times bigger

To reconstruct one failed drive in RAID 5 or RAID 6, every surviving drive is read essentially end to end.

Six 8 TB drives: you read 40 TB to write 8.

RAID 10 doesn't work that way at all. A failed drive is rebuilt by copying its mirror partner — one drive, 8 TB, not 40. That structural difference is the strongest practical argument for mirroring on large arrays, and it's completely invisible in any comparison based on usable capacity.

Drives publish an error rate

It's on the datasheet, under non-recoverable read errors per bits read.

Drive class	Spec	Data per allowed error
Consumer / NAS — IronWolf, SkyHawk	1 per 10¹⁴ bits	12.5 TB
Enterprise — Exos and similar	1 per 10¹⁵ bits	125 TB

12.5 TB is the number worth remembering. On a consumer spec you're entitled to expect one unreadable sector for roughly every 12.5 TB read — and a six-drive rebuild reads 40.

That's where the alarming arithmetic comes from: at the spec maximum, the default array expects 3.2 read errors during a single rebuild, giving a 95.9% chance of at least one.

Now the honest part: that's a ceiling, not a forecast

The well-known "RAID 5 is dead" argument runs on exactly this calculation, and it overreaches.

The published figure is a maximum the manufacturer guarantees not to exceed — not the rate drives actually achieve. Measured error rates are generally far better than spec, and real rebuilds succeed much more often than this predicts.

What the number is genuinely good for is comparison: between levels, between drive classes, between array sizes. Those conclusions hold even though the absolute probability is pessimistic.

What double parity actually buys

Here's where it matters, and where a single probability figure misleads.

RAID 5 and RAID 6 read the same drives, so their chance of hitting a read error is identical. What differs is whether the array can absorb it — RAID 6 still has a second parity while rebuilding the first failure.

Setup	At least one error	Level can't absorb it
RAID 5, consumer	95.9%	95.9%
RAID 6, consumer	95.9%	82.8%
RAID 6, enterprise	27.4%	4.1%

That last row is the practical answer for a large array: double parity and enterprise drives together take the spec-max risk from near-certain to a few percent.

(Even the RAID 6 figure is pessimistic twice over — two errors would also have to land in the same stripe to be unrecoverable.)

It scales badly, which is the real point

Twelve 20 TB drives in RAID 5:

• 220 TB read to rebuild one drive
• 17.6 expected errors at spec maximum
• 55.6 hours of rebuild at 100 MB/s

Drives keep growing, so rebuild windows keep lengthening while the amount that must be read grows with them. That is why single parity fell out of favour on large arrays — more than any individual probability figure.

One read error is not always a dead array

Older hardware controllers would fail an entire rebuild on a single unreadable sector, which is where the scariest version of this story comes from.

Modern systems are often far more graceful. ZFS and btrfs will report which file was affected, carry on rebuilding, and leave everything else intact.

Losing one file is a completely different day from losing the array — and which one you get is a property of your controller and filesystem, not of this arithmetic. Worth knowing which you have before you need to.

What to actually do when a drive dies

Verify your backup before starting the rebuild. This is the single riskiest window your array will ever be in — the moment it has the least redundancy it will ever have while still working.

Then: check the other drives' SMART data, because a second failure often isn't far behind; prioritise rebuild speed over array performance if your controller allows it; and keep the array as idle as you can for the duration.

The rebuild window is the risk. Everything that shortens it helps.

---

Sources: drive error-rate specifications from Seagate IronWolf Pro and Seagate Exos X10 product documentation.