Storage Overhead Calculator

Usable capacity is what survives a chain of multiplicative deductions, which is why the losses compound rather than add. Raw capacity is first multiplied by the RAID ratio, which is (N-1)/N for single parity, (N-2)/N for double parity and 0.5 for mirroring. The survivor is then reduced by filesystem overhead, and what remains is reduced again by the snapshot reserve. Applying 5 percent and 15 percent in sequence removes 19.25 percent, not 20 percent. Filesystem overhead covers superblocks, inode tables, allocation bitmaps and journals, and varies by design: ext4 reserves 5 percent for the root user by default and allocates its inode table at format time, NTFS sets aside a zone for the Master File Table, XFS metadata typically costs 1 to 3 percent, and ZFS is usually planned with 10 to 20 percent headroom because copy-on-write allocation degrades badly on a full pool. Snapshot reserve sizing follows the change rate of the data, since a snapshot only consumes space for blocks that are subsequently modified. Archive and file shares usually need 5 to 10 percent, database volumes 15 to 20 percent, and virtual machine datastores 25 to 30 percent. Exhausting the reserve is not a soft failure: depending on the platform the array will either silently expire old snapshots or reject writes. Three deductions are not modelled here and should be subtracted separately: dedicated hot spares, which remove whole drives from the pool; unit conversion, since a 10 TB decimal array presents as about 9.09 TiB to an operating system that divides by 1024; and any thin-provisioning oversubscription, which lets allocated capacity exceed the real usable figure computed above.

Reserving part of a disk from users predates RAID entirely. The Berkeley Fast File System, described by Marshall Kirk McKusick, William Joy, Samuel Leffler and Robert Fabry in 1984, organized disks into cylinder groups and deliberately held back a free-space reserve, originally 10 percent, because block allocation and layout quality collapse once a volume approaches full. That reserve, later reduced to 5 percent, was inherited by ext2 when Remy Card introduced it for Linux in 1993 and survives as the default in ext4 today. Microsoft made a parallel choice when NTFS shipped with Windows NT 3.1 in 1993, keeping a contiguous zone reserved for the Master File Table so that metadata would not fragment across the volume. The snapshot layer was added by network storage vendors in the 1990s. NetApp, founded by Dave Hitz and James Lau, built the WAFL filesystem around copy-on-write snapshots and shipped volumes with a default snapshot reserve, historically 20 percent, making the reserve an explicit line item in every capacity plan. Sun Microsystems generalized the same ideas in ZFS, released in 2005, where snapshots, checksums and copy-on-write allocation all draw from a shared pool and administrators are advised to keep utilization below roughly 80 percent. The final component, the decimal-versus-binary gap, was formally resolved when the International Electrotechnical Commission defined the tebibyte and its relatives in a 1999 amendment to IEC 60027-2, though the roughly 9 percent discrepancy still surprises buyers comparing a purchase order against a mount point.