Unraid Storage & Parity Rebuild Calculator

How the parity maths works

Three rules decide every figure above, and none of them needs a spreadsheet once you have seen it written out.

The minimum parity drive size rule

Unraid does not stripe data, so a parity disk has to cover the highest block offset that exists on any single data disk. That makes the rule absolute: no data disk may be larger than the smallest parity disk. A 12 TB parity disk caps every data disk in the array at 12 TB, and the only way to add a 16 TB disk later is to replace parity first and rebuild it. Because parity is fixed at purchase time rather than adjusted afterward, the practical move is to buy the largest drive in the build as parity, even when the data disks are smaller or second-hand. With dual parity, both parity disks must clear the largest data disk, but they do not have to match each other.

Dual parity and fault tolerance

A single parity disk reconstructs exactly one missing data disk. A second parity disk uses a different calculation, so the array can lose any two disks at once and still rebuild both. The number worth watching is not the array size but the rebuild window: reconstructing a large disk reads every other disk in the array from end to end for hours, and that sustained load is when a marginal second disk is most likely to drop. Once an array holds more than about six drives, the chance of a second failure during that window is high enough that dual parity becomes the standard recommendation. Dual parity costs the capacity of two drives whatever the array size, so its overhead as a percentage falls as the array grows.

Estimating parity sync and rebuild duration

A parity sync and a disk rebuild take roughly the same time, because both move one full parity disk end to end. The arithmetic is capacity divided by sustained transfer rate: an 18 TB drive at a steady 200 MB/s works out to 18,000,000 MB ÷ 200 MB/s ≈ 90,000 seconds, about 25 hours. Halve the speed and the time doubles; the drive's own throughput, not the array's disk count, sets the floor. Real runs usually land slower than the raw division suggests, because a check or rebuild competes with normal array reads and writes and platter drives slow toward their inner tracks. Treat the figure above the slider as a best case and leave headroom in the maintenance window.