RAID Calculator — Usable Capacity by Level

A RAID array trades raw capacity for fault tolerance, and each level sets that exchange rate differently. Striping alone (RAID 0) keeps all N disks of capacity but tolerates zero failures. Mirroring (RAID 1) yields the capacity of one disk while surviving the loss of all but one member. Single distributed parity (RAID 5) gives (N-1) disks of usable space and survives one failure; double parity (RAID 6) gives (N-2) and survives two; striped mirrors (RAID 10) give N/2 and reliably survive one failure, or more if the failures fall in different mirror pairs. Storage efficiency is simply usable divided by raw, so RAID 5 across 8 disks is 87.5 percent efficient while the same disks in RAID 10 are 50 percent. Minimum member counts are 2 for RAID 0 and 1, 3 for RAID 5, and 4 for RAID 6 and RAID 10, and in a mixed array every disk is truncated to the size of the smallest member. The critical limitation is what happens during a rebuild. After a failure the array runs degraded and every surviving disk is read end to end, which can take 6 to 24 hours or more on multi-terabyte drives and raises the chance of a second failure exactly when there is no redundancy left. Consumer SATA drives are typically specified at one unrecoverable read error per 10 to the 14th bits, roughly one per 12.5 TB read, which is why single parity is discouraged on large arrays and RAID 6 or RAID 10 is standard in production. RAID is also not a backup: it protects against disk failure, not deletion, ransomware, corruption, controller failure or site loss.

The concept and the acronym come from a 1988 SIGMOD paper, A Case for Redundant Arrays of Inexpensive Disks, by David Patterson, Garth Gibson and Randy Katz at the University of California, Berkeley. Their argument was economic: rather than buying a single large expensive mainframe disk, an array of commodity drives could deliver more capacity and throughput for less money, provided redundancy compensated for the higher combined failure rate. The paper defined the original levels 1 through 5, with level 5 distributing parity across all members to avoid the dedicated-parity-disk bottleneck of level 4. Level 0 and the nested combinations such as RAID 10 were named afterwards by industry rather than in the paper, and vendors gradually re-read the acronym as Redundant Array of Independent Disks once the drives ceased to be notably inexpensive. The RAID Advisory Board, formed in the early 1990s, standardized terminology as controllers proliferated, and double-parity RAID 6 entered common use as drive capacities grew and single-parity rebuilds became statistically risky. That risk was publicly argued through the late 2000s, most visibly in Robin Harris 2007 essay predicting that RAID 5 would become untenable as capacity outran unrecoverable read error rates, and the industry did move toward RAID 6 and mirroring for large arrays. Sun Microsystems addressed a related weakness, the parity write hole, with RAID-Z in ZFS, released in 2005, which uses variable-width full-stripe writes so a partial stripe update can never leave parity inconsistent.