RAID calculator

This tool supports RAID 0, 1, 4, 5, 6, and 10, returns usable_capacity in the same unit you used for disk_size, and reports fault_tolerance as a disk count. If disk_count is below the level's minimum the task fails with invalid_input instead of inventing a layout that no controller would assemble. The free browser path and the prepaid API share one deterministic module; successful API calls cost $0.002.

RAID formulas: usable capacity, overhead, and fault tolerance

RAID usable capacity is never a mystery once the disks are equal. RAID 4 keeps a dedicated parity disk; RAID 5 rotates that single parity stripe across members. Two failures in the same pair still kill that stripe. Nested RAID 50 and RAID 60 need a group count this three-field contract does not accept, so they are not offered. Raw capacity is disk_count times disk_size. RAID 0 stripes every block across all members with no copy and no parity, so usable_capacity equals raw capacity and fault_tolerance is zero: any single disk failure loses the array. RAID 1 is an n-way mirror: usable_capacity equals one disk_size, and the array survives until the last remaining disk fails, so fault_tolerance is disk_count minus one. Both need at least three disks, both give usable_capacity of (disk_count minus one) times disk_size, and both tolerate exactly one disk failure. RAID 6 stores two independent parity syndromes, needs at least four disks, gives usable_capacity of (disk_count minus two) times disk_size, and tolerates any two disk failures. RAID 10 is a stripe of two-disk mirrors: disk_count must be even and at least four, usable_capacity is half of raw, and one disk may fail in each pair (up to disk_count over two). Efficiency_percent is usable_capacity divided by raw_capacity times one hundred, so RAID 5 with four disks is seventy-five percent and RAID 10 is always fifty. These identities assume identical members; mixed sizes are truncated to the smallest disk by real controllers and are out of scope here. Overhead is simply raw_capacity minus usable_capacity: the space spent on copies or parity rather than host data.

Input contract, level minima, and the response shape

RAID 10 additionally requires an even count so members pair into mirrors. None of those failures are billed. Send a JSON body with three required fields. Field raid_level is the closed set 0, 1, 4, 5, 6, or 10; aliases level, raid, and raid_type are accepted, and a leading raid prefix is stripped so RAID-5 and 5 mean the same layout. Field disk_count is a positive integer from two through two hundred fifty-six; aliases disks, n, drives, and drive_count map to it. Field disk_size is a positive finite capacity in any consistent unit—gigabytes, terabytes, or tebibytes—because every output capacity is returned in that same unit; aliases size, capacity, drive_size, and disk_capacity are accepted. Minimum disk_count depends on the level: two for RAID 0 and RAID 1, three for RAID 4 and RAID 5, four for RAID 6 and RAID 10. A disk_count below the level minimum is invalid_input naming disk_count; an odd RAID 10 count is invalid_input for the same field; a missing or non-finite disk_size is invalid_input naming disk_size. On success the body echoes raid_level, disk_count, and disk_size; names the layout (raid_name, scheme); lists min_disks, data_disks, and redundancy_disks; and returns raw_capacity, usable_capacity, overhead, efficiency_percent, fault_tolerance, fault_tolerance_note, formula, and a one-line message. Example: raid_level 5, disk_count 4, disk_size 2 yields usable_capacity 6, overhead 2, efficiency 75, and fault_tolerance 1. Pricing is a flat $0.002 per successful request with unit item.

Planning arrays before you buy disks

Homelab NAS shoppers comparing consumer RAID 5 against RAID 6 should weigh the extra disk of RAID 6 against the dual-parity rebuild window, which this tool does not simulate: it reports static capacity and static fault_tolerance, not rebuild time or unrecoverable-read-error risk during a rebuild. Keep disk_count at or below two hundred fifty-six and disk_size inside the published positive range so intermediate products stay finite in IEEE double precision on the Worker and in the browser. Failed validation does not consume prepaid balance, which keeps exploratory scripts cheap when you try RAID 5 with two disks and learn the minimum is three. Successful API calls cost $0.002 per request.