Six-Bay NAS Drives: The Rebuild Math at 20TB+

Twenty-four terabytes takes the better part of a day to copy at the drive maker’s own best-case number. A rebuild is messier than that. And the rebuild is exactly when you find out whether you bought the right layout.

The NAS hard drive guide covers which drives are good. This page answers the next question: which 20TB-plus drives belong in a six-bay box once rebuild time and failure behaviour are part of the price?

Some retailer links below are affiliate links; if you buy through them TechFuel HQ may earn a small commission at no extra cost to you. As an Amazon Associate I earn from qualifying purchases. Commissions never influence what gets recommended — see our disclosure.

Bigger is not reliably cheaper per terabyte

Nine exact model numbers, lowest in-stock price each, August 2026:

Four of the six capacity steps lower cost per terabyte. Two raise it. WD Red Pro improves from 20TB to 22TB then gets worse at 24TB. Seagate gets worse at 22TB and recovers at 24TB. Toshiba falls at both steps.

The “buy the biggest drive, it’s cheaper per TB” rule of thumb is wrong often enough to be worth thirty seconds with a calculator. Six drives from this grid runs $4,752 to $6,360, so a step in the wrong direction is real money.

Prices move. Check before you buy six of anything.

The rebuild floor

Take the most generous possible assumption — the drive sustains its rated maximum across the whole platter, the replacement is empty, nothing else is using the NAS — and a full copy still takes this long:

Rates from the Seagate Exos X24 manual, the WD Red Pro data sheet, and the Toshiba MG11 overview.

You will not hit those numbers. TrueNAS’s dRAID primer explains why: a traditional RAIDZ resilver reads every surviving disk at once and chases scattered blocks, which turns much of the job into random I/O rather than the clean sequential streaming the spec sheet describes. Add parity work, whatever else the NAS is serving, and how full the pool is.

Six-wide at 24TB means reading up to 120TB and writing 24TB — 144TB of device traffic through one rebuild. Plan for a day. The three drives are close enough on paper that none of them buys you a meaningfully shorter window.

The URE calculation is the wrong reason to pick RAIDZ2

You will see this argument everywhere, so it is worth dismantling properly.

All three current 24TB drives are rated around one unrecoverable sector per 10^15 bits. The famous calculation treats that as an independent per-bit probability:

P(at least one error) ≈ 1 − exp(−B/q)

Five full 24TB survivors hold 960 trillion bits. At q = 10^15 that returns 61.7%. Using the older consumer-grade 10^14 figure returns 99.99% — which is where “RAID 5 is dead” came from.

The arithmetic is fine. The assumption underneath it is not. A qualification limit is a bound the manufacturer guarantees, not a claim that every bit is an identical coin flip at exactly that rate.

The field data disagrees with the model directly. Bairavasundaram et al. studied 1.5 million drives over 32 months (FAST/SIGMETRICS, 2007) and found latent sector errors in 3.45% of nearline drives and 1.46% of enterprise drives — with strong temporal and spatial clustering, and five of eighteen drive families recording none at all. Errors bunch up in particular drives and particular families. They do not sprinkle themselves evenly across your array.

That study does not prove RAIDZ1 is safe. It proves 61.7% is not a real prediction about your rebuild.

So pick redundancy on structure, not on that percentage — which lands you in the same place for a better reason.

What to actually build

OpenZFS gives RAIDZ1 one device of fault tolerance and RAIDZ2 two. With six 24TB drives:

Build one six-wide RAIDZ2 vdev. You give up 24TB to keep one device of fault tolerance through a copy job whose floor is already 21 hours. That is the entire argument, and it does not need the disputed percentage to work.

Choose three mirrored pairs instead if you want faster random I/O, want to expand two drives at a time, and would rather a rebuild read one disk than five. The sharp edge: lose both members of one pair and the pool is gone while four healthy drives watch.

Do not build six-wide RAIDZ1 at these capacities for anything you cannot restore from somewhere else. It saves one drive and spends every bit of your failure margin during the longest maintenance window the array will ever have.

The NAS capacity calculator runs the same parity maths for other drive counts and sizes.

Buy the Exos X24

Six Seagate Exos X24 24TB in one RAIDZ2 vdev. $5,688 for 96TB usable.

It is the cheapest 24TB drive in the class at $39.50/TB — $112 a drive under the WD Red Pro, which is $672 more across six bays for a disk that rebuilds no faster. Exos is also the cheapest option at 20TB, so if you want a smaller pool the answer does not change: six 20TB Exos at $791.99 is $4,752 for 80TB usable.

The rebuild argument does not separate these drives — 23.4 hours against 23.2 and 21.6 is noise once parity and live traffic are in play. So the money decides, and the money is not close.

Two things that would change the pick. If your enclosure’s compatibility list names Red Pro and not Exos, follow the list — a supported drive beats a cheaper one. And if you are buying into an existing pool, match what is already in the bays rather than mixing families.

RAIDZ2 survives a drive dying. It does not survive you deleting the wrong directory — keep a backup somewhere else.