RAID Calculator

RAID capacity formulas

Let N be the number of active member disks, S the smallest active-disk capacity, P the sum of their actual capacities and G the number of equal spans in RAID 50 or 60. Conventional RAID limits each member to S, so the RAID-addressable pool is N × S; mixed-size waste is P − N × S.

RAID 50 needs at least two equal spans of three disks; RAID 60 needs at least two equal spans of four. The conditional maximum is not a promise that any combination of that many failures survives. Two failed members of one RAID 10 pair lose that pair; too many failures in one parity span lose the nested array.

Worked mixed-drive example

Five active drives of 8, 8, 6, 6 and 6 TB in RAID 6 contain 34 TB physically. The 6 TB smallest drive limits the addressable pool to 5 × 6 = 30 TB, leaving 4 TB unused on the larger disks. Double-parity overhead is 12 TB, so theoretical usable capacity is 18 TB, or about 16.371 TiB. Controller metadata, system partitions, pools and filesystems will reduce the volume presented to applications further.

TB is not TiB

Disk manufacturers normally use decimal units: 1 TB = 1,000,000,000,000 bytes. The IEC binary prefix is 1 TiB = 2⁴⁰ = 1,099,511,627,776 bytes. They describe the same bytes with different scales. Four 4 TB drives in RAID 5 therefore provide 12 TB but about 10.914 TiB, not “12 TB that shrank.” The calculator converts through bytes whenever the input unit changes.

Hot spares, rebuilds and performance

A hot spare waits outside the active layout. It adds no usable capacity and does not raise the array’s immediate simultaneous-failure tolerance. After an active disk fails, a compatible spare may be assigned automatically and rebuilt, shortening time spent degraded; redundancy returns only when that rebuild succeeds. This calculator checks that spare capacity meets the modeled member capacity, but controllers may require the same or a larger physical drive, compatible media or interface, and other vendor-specific eligibility rules. The controller’s rules prevail. Actual rebuild speed depends on drive size and type, controller settings, stripe and cache policies, enclosure and live workload, so this calculator does not invent a completion time.

Performance is similarly workload-specific. Striping may improve throughput; mirrors may serve reads from multiple copies; parity layouts add write work. RAID 10 avoids parity calculations, while RAID 5/6/50/60 must maintain parity. None of those facts justifies an N× speed promise without measurement on the real controller and workload.

RAID availability is not backup. It does not protect against deletion, corruption, malware, controller failure, fire or theft. Keep independent, tested backups.

Primary references: SNIA’s RAID 10 definition, IBM’s RAID capacity formulas and smallest-drive rule, Lenovo’s RAID level, span and hot-spare guide, Dell PERC RAID level requirements, QNAP’s mixed-capacity example, NIST’s binary prefixes, Seagate’s decimal/binary capacity explanation, and Dell’s documented rebuild-rate factors.