RAID capacity formulas
Below, n is the number of active disks (excluding spares), s the size of each disk and g the number of span groups in RAID 50/60.
RAID 0 : usable = n × s tolerance 0 (min 2 disks)
RAID 1 : usable = s tolerance n − 1 (min 2)
RAID 5 : usable = (n − 1) × s tolerance 1 (min 3)
RAID 6 : usable = (n − 2) × s tolerance 2 (min 4)
RAID 10 : usable = (n ÷ 2) × s tolerance 1 (up to n/2) (min 4, even)
RAID 50 : usable = (n − g) × s tolerance 1 per group (min 3 per group)
RAID 60 : usable = (n − 2g) × s tolerance 2 per group (min 4 per group)
efficiency = usable ÷ (n × s) 1 TB = 10¹² B = 0.9095 TiB
Rebuild risk and URE
When a disk fails, the controller must read every surviving disk in the group in full to rebuild the replacement. Vendors specify an unrecoverable read error rate; the chance of hitting at least one during that read is approximately:
bits_read = data read during rebuild (TB) × 8 × 10¹²
P(≥1 URE) = 1 − e^(−bits_read ÷ URE)
rebuild_time ≈ disk size ÷ rebuild speed
In RAID 5 such an error during a rebuild can lose some or all data; RAID 6 still has a parity layer to correct it. The model is conservative — the URE spec is an upper bound and healthy disks usually do better — but it is very useful for comparing levels and deciding about large disks.
Selection tips
- RAID 10 gives the best performance for databases and random writes (write penalty 2).
- RAID 6 balances space and safety for archives and file servers on large disks.
- RAID 50/60 shorten rebuilds on big arrays by keeping each group small.
- On large arrays keep at least one hot spare so a rebuild starts immediately.