RAID Capacity Calculator

Usable RAID capacity is raw capacity minus whatever the array spends on redundancy. Striping alone (RAID 0) spends nothing, so N disks of size S give N times S but tolerate no failure. Mirroring (RAID 1) writes every block twice, so a two-disk mirror yields S. Single parity (RAID 5) reserves one disk worth of parity across the stripe, giving (N-1) times S; dual parity (RAID 6) reserves two, giving (N-2) times S; and a striped mirror (RAID 10) yields N/2 times S. Efficiency rises with member count for parity levels while fault tolerance does not: a 12-disk RAID 5 is 92 percent efficient yet survives one failure. Minimum members are 2 for RAID 0 and 1, 3 for RAID 5, and 4 for RAID 6 and 10. Several caveats separate this arithmetic from delivered capacity. Drive vendors quote decimal terabytes while operating systems usually report binary tebibytes, so a 2 TB disk shows as roughly 1.82 TiB, an immediate 9 percent shortfall before any filesystem overhead, and arrays are sized by their smallest member. The more serious limit is rebuild risk: with consumer drives quoting an unrecoverable read error rate near one sector in 10 to the 14th bits, a full rebuild of a wide array carries a real chance of a second fault, which is why RAID 6 or mirroring is preferred once disks pass a few terabytes. Parity arrays also suffer the write hole on unclean shutdown unless protected by a journal, battery-backed cache, or a design such as ZFS RAID-Z. RAID protects against drive failure only; it is not a backup, since deletion, corruption, and ransomware propagate to every member at once.

The RAID taxonomy comes from a single 1988 paper. David Patterson, Garth Gibson, and Randy Katz of the University of California, Berkeley presented A Case for Redundant Arrays of Inexpensive Disks at the ACM SIGMOD conference, arguing that arrays of small commodity drives could beat the large, expensive single disks then sold for mainframes on cost per megabyte, provided redundancy compensated for the much shorter expected life of the array. That paper defined levels 1 through 5, distinguishing mirroring, bit-level and block-level striping with dedicated parity, and the distributed-parity arrangement that became RAID 5. The industry later reread the acronym as Redundant Array of Independent Disks. Levels the paper did not define were added by practice and by vendors. Pure striping with no redundancy was labelled RAID 0 after the fact, nested combinations such as RAID 10 and RAID 50 came from controller vendors, and dual-parity RAID 6, which uses a second syndrome computed with Reed-Solomon coding to survive two simultaneous failures, was standardized later as capacities grew and rebuild windows lengthened. More recent designs move away from fixed levels entirely: ZFS RAID-Z, released with Solaris in 2005, uses variable-width stripes to close the write hole, and declustered or distributed parity schemes in modern storage systems spread rebuild work across many drives to shorten the exposure window that motivated RAID 6 in the first place.