RAID trades disks for capacity, speed and survival in fixed exchange rates — and the rates differ enough by level that guessing costs terabytes. This RAID Calculator takes the number of disks, the size of the smallest disk and the RAID level, and returns the array’s usable capacity, its space efficiency and the disk failures it can survive.
Worked Example: Four 4 TB Disks, Every Level
Sixteen raw terabytes enter; each level keeps a different share. RAID 0 keeps all of it — 4 × 4 = 16 TB at 100% efficiency — and survives zero failures: lose any disk, lose the array. RAID 1 mirrors: usable capacity equals one disk, 4 TB (25% efficiency at four disks), surviving up to three failures if the mirrors hold — the most protection per array and the worst capacity economics. RAID 5 sets aside one disk’s worth of parity: (4 − 1) × 4 = 12 TB at 75%, surviving any single failure. RAID 6 sets aside two: (4 − 2) × 4 = 8 TB at 50%, surviving any two failures. RAID 10 mirrors pairs, then stripes: (4 / 2) × 4 = 8 TB at 50%, surviving at least one failure per mirrored pair — possibly two, if fate picks different pairs. The table is the whole decision in miniature: capacity efficiency rises as protection thins, and the right level is the one whose failure column matches your recovery plan — not a property of the disks. One unit discipline before ordering: disk makers sell decimal terabytes (1 TB = 10¹² bytes) while operating systems report binary tebibytes, so the array you build as “12 TB” mounts as about 10.9 TiB — both are the same bytes wearing different labels, and the calculator reports the maker’s convention your invoice used.
Parity, Rebuilds and the RAID 5 Argument
RAID 5’s single parity is information theory in work clothes: for each stripe, one block holds the XOR of the others, so any one missing block can be reconstructed from the survivors — the same XOR trick this site’s Parity Bit Calculator demonstrates bit by bit. The level’s controversy is not the math but the rebuild: when a disk dies, the array must read every remaining disk end to end to reconstruct the lost one, and on modern multi-terabyte disks that read takes many hours, during which a second failure (or an unrecoverable read error on a survivor) destroys the array. That exposure window is why large-disk practice has moved to RAID 6’s double parity (two independent parity blocks, surviving any two failures, including one mid-rebuild) and why hot spares — an idle disk the controller rebuilds onto automatically — are standard in serious arrays. Two further honest notes: write performance on parity RAID pays a read-modify-write tax on small writes (RAID 10’s mirroring avoids it, one reason databases love 10), and no level on this page protects against deletion, corruption, ransomware, fire or theft — RAID is availability, not backup, a sentence embroidered on the wall of every sysadmin who has learned it the other way. The calculator sizes the array; the backup plan is a separate, non-negotiable document.
Sizing Rules the Calculator Assumes
Four conventions govern the arithmetic, stated so your array and the page agree. Smallest disk rules: arrays build on the smallest member — a 4 TB disk among 8 TB disks contributes 4 TB to every stripe in classic implementations, which is why the input says “smallest disk if mixed.” Minimum counts: RAID 0 needs 2+, RAID 1 needs 2, RAID 5 needs 3, RAID 6 needs 4, RAID 10 needs 4 and an even count — the calculator enforces these floors and says so rather than printing fantasy arrays. Efficiency is capacity-only: performance (IOPS, rebuild speed, write penalties) is a different model per workload; this page prices space and survival, the two things a purchase order must state. Beyond the five classics: RAID 50/60 nest parity groups under a stripe and scale the same formulas per group; ZFS RAID-Z and vendor schemes (SHR and relatives) vary the details — notably letting mixed disk sizes contribute more of their capacity — while preserving the parity idea. And a sizing habit that outlives every level debate: leave headroom. Arrays that run at 95% full rebuild slowly, fragment badly and leave nowhere for a failed disk’s data to visit — design to 80% of the usable figure the calculator gives you, and the array will age like infrastructure instead of like a countdown. One last counting rule: a hot spare does not belong in the disk count you enter here. It holds no data and joins no stripes while it idles — it is the understudy, not a cast member — so enter the array’s working members, and let the spare’s existence comfort the rebuild arithmetic of the earlier section instead of inflating the capacity.
Write Penalties: The Other Exchange Rate
Capacity is only half the pricing; every level also taxes writes, in extra disk operations per logical write, and the table explains a decade of purchasing decisions. RAID 0 and RAID 1/10 pay the minimum — a mirror write costs 2 I/Os (one per copy), a stripe costs 1. RAID 5 pays 4: a small write must read the old data and old parity, then write the new data and new parity. RAID 6 pays 6, with its second parity block joining the read-modify-write dance. The consequence is workload-shaped: sequential and read-heavy work barely notices parity’s tax (full-stripe writes skip the reads entirely), while small random writes — databases, virtual-machine stores, mail spools — feel it at full strength, which is why those workloads live on RAID 10 despite its 50% capacity efficiency. Battery- or flash-backed write caches exist to absorb exactly this penalty, converting many small writes into full-stripe flushes. The calculator prices capacity and survival; when you apply its answer to a write-heavy service, read this section as the second half of the quote and choose the level whose penalty your workload can afford.
The URE Argument, Stated Carefully
The strongest version of the case against large-disk RAID 5 runs on one statistic: consumer drives are rated around one unrecoverable read error (URE) per 10¹⁴ bits read. Rebuilding a failed member of a 12 TB-usable array means reading the survivors’ roughly 12 TB — about 9.6 × 10¹³ bits — so the expected number of UREs encountered during that single rebuild is of order one. Not a certainty — an expectation value, which across a fleet and a career behaves like one — but it is the arithmetic that moved enterprise practice to RAID 6, dual-parity ZFS and erasure coding as disks grew: the rebuild of a single-parity array has become a coin weighted by terabytes. The honest counterweights: enterprise drives rate URE at 10¹⁵ bits (ten times better), real-world error rates are debated, and a URE during rebuild typically costs one stripe’s file, not the array, on modern systems that can isolate it. None of that rescues complacency — it relocates the conclusion: for multi-terabyte members, price the second parity disk (this page’s RAID 6 column) as insurance whose premium is known and whose alternative is a rebuild conducted under statistical fire. And hold the line that no parity geometry of any width substitutes for the backup.
Common Mistakes
Treating RAID as a backup and skipping the off-array (ideally off-site) copy. Mixing disk sizes and expecting the large disks’ full capacity to count. Choosing RAID 5 for very large disks without pricing the rebuild-risk argument for RAID 6. Assuming RAID 10 survives any two failures (it survives two only if they land in different mirror pairs). Forgetting the TB-versus-TiB relabeling and reporting a “missing” 9% to the vendor. And filling the array to the brim on day one, leaving no headroom for rebuilds, snapshots or growth.
Related Tools
Continue with the Upload Time Calculator, the Parity Bit Calculator, the CIDR Calculator, or browse the full Other Calculators hub.
Frequently Asked Questions
How much usable space does RAID 5 give?
(Number of disks − 1) × smallest disk size. Four 4 TB disks give (4 − 1) × 4 = 12 TB usable, at 75% efficiency, surviving any one disk failure.
What is the difference between RAID 5 and RAID 6?
One parity block versus two. RAID 6 sets aside two disks’ worth of parity, surviving any two simultaneous failures — the safer choice for large disks whose long rebuild windows leave a single-parity array exposed.
Is RAID a backup?
No. RAID protects against disk failure so the system stays available; it happily mirrors deletions, corruption and ransomware in real time. Backups — versioned, separated, ideally off-site — are a separate discipline.
Why does my 12 TB array show as about 10.9 TB?
Units: disks are sold in decimal terabytes (10¹² bytes) and your OS reports binary tebibytes (2⁴⁰). 12 × 10¹² / 2⁴⁰ ≈ 10.9 TiB — the same bytes, relabeled.
Can I mix disk sizes in RAID?
Classic RAID sizes every stripe to the smallest disk, so larger disks donate their surplus to waste. Some modern systems (ZFS variants, vendor hybrid RAID) use more of the big disks — check your platform before buying mixed sizes.
How many disks does RAID 10 need?
At least four, and always an even number — disks form mirrored pairs that are then striped. Usable capacity is half the raw total, and the array survives one failure per pair.