RAID Levels Explained: RAID 0, 1, 5, 6, and 10 Compared

RAID — redundant array of independent disks — spreads data across several drives so that one drive’s failure does not lose everything, or so that reads and writes go faster, or both. Each level makes a different trade between usable capacity, performance, and fault tolerance, and the right level depends on which of those three you need most. This guide compares RAID 0, 1, 5, 6, 10, and 0+1 on the same disks, explains parity and mirrors in plain words, and covers the rebuild window, when protection is thinnest.

What RAID is, and the three dials it trades

A RAID array presents several physical disks as one volume, with the controller or software deciding how data is laid out. Striping splits data across disks so reads and writes happen in parallel, which is fast. Mirroring writes the same data to two disks, so either one can fail without loss. Parity writes error-correction information across the disks, so a failed disk’s contents can be recomputed from the survivors. Every level is a combination of these three ideas, and every level trades capacity, behaviour, and failure tolerance against each other. This calculator prices only the first — capacity — and this guide prices all three, so the choice is made with the full trade in view.

The three dials are separate. Capacity is how much usable space the array gives. Behaviour is how it reads, writes, and rebuilds — striped reads are fast, parity writes carry a penalty, mirrors read fast and write at single-disk speed. Failure tolerance is how many disks can fail, and in which combinations, before data is lost. When choosing between levels, price all three: what the array holds, what it survives, and how it behaves in the hours after something has already gone wrong. A level that wins on capacity and loses on rebuild behaviour is a bad choice for data you cannot afford to lose twice.

Every level on the same disks

Take four 4 TB disks, 16 TB raw. RAID 0 stripes across all four: 16 TB usable, no redundancy, and any single disk failure loses the array. It exists for speed and for scratch space where loss does not matter. RAID 1 mirrors: with four disks the usual layout is two mirrored pairs or a four-way mirror, and the usable space is one disk, 4 TB — the most expensive protection per terabyte, and the simplest to reason about.

RAID 5 distributes one disk’s worth of parity across the array: (4 − 1) × 4 TB = 12 TB usable, surviving any single disk failure. RAID 6 distributes two disks’ worth: (4 − 2) × 4 TB = 8 TB usable, surviving any two failures. RAID 10 mirrors pairs and then stripes across them: 8 TB usable, surviving at least one failure, and a second if it falls in the other pair. RAID 0+1 stripes first and mirrors second: the same 8 TB and the same headline tolerance, but a different failure geometry, covered below. The RAID Calculator applies these rules to any disk count and size, and shows the raw total, the space handed to protection, the efficiency, and the failure count alongside.

RAID 6 needs 4 disks to be useful in practice; the table shows the 4-disk comparison throughout.

Parity and mirrors, in plain words

Mirroring is duplication. Every byte is written twice, to two disks, so either disk alone holds the whole array. Reads can come from either disk, which helps read speed; writes must complete on both, which caps write speed at one disk. A mirror rebuild copies one disk to its replacement, reading only the surviving mirror — fast, and gentle on the array. The cost is capacity: half the raw space is a copy.

Parity is arithmetic. For RAID 5, the parity block for a stripe is the XOR of the data blocks: the parity of 1, 0, 1 is 0, and any one missing block can be recomputed from the rest. The parity is rotated across disks so no single disk holds all of it. Writes cost more, because changing one data block requires reading the old data and parity, computing the new parity, and writing both back — the read-modify-write penalty. RAID 6 uses two independent parity calculations, which is why it survives two failures and why its write penalty is larger.

Efficiency improves with array size for parity RAID, and the protection does not change. RAID 5 on four 4 TB disks gives 12 TB at 75.0% efficiency; on six 6 TB disks it gives 30 TB at 83.3%; on eight 8 TB disks it gives 56 TB at 87.5%. The protection — surviving any single disk — is the same in every case; only the overhead share shrinks as the array grows. That scaling is why RAID 5 is attractive for large arrays, and why its rebuild problem, next, grows with it.

RAID 10 and RAID 0+1: same capacity, different survival

RAID 10 and RAID 0+1 give the same capacity and look identical in a specification table, but they fail differently. RAID 10 mirrors first: disks A and B mirror each other, C and D mirror each other, and data is striped across the two pairs. Lose A, and B still holds that half; lose C as well, and D still holds the other half — the array survives two failures, provided they are in different pairs. Lose A and B together, and that half is gone.

RAID 0+1 stripes first: A and B form a stripe, C and D form an identical stripe, and the two stripes mirror each other. Lose A, and the whole A+B stripe is gone, but the C+D mirror still holds everything. Lose C as well — one disk from each stripe — and the array is lost, because neither stripe is whole. RAID 10 therefore survives more two-failure combinations than RAID 0+1, and its rebuilds are faster, because a rebuild reads only the surviving mirror of the failed disk rather than a whole stripe. Given the choice, RAID 10 is the better nested level in almost every case.

Two levels that do not exist deserve a word. There is no standard RAID 7 — the name was a proprietary product, not a level — and there is no RAID 2, 3, or 4 in modern use: RAID 2 used Hamming codes, RAID 3 and 4 used a dedicated parity disk, and all three were superseded by distributed parity in RAID 5 and 6. If a specification sheet offers RAID 7, it is selling something else under a familiar name, and the sheet should be read as a product description, not a standard.

Rebuilds: the window when protection is thinnest, and why RAID is not a backup

When a disk fails, the array is degraded: it still serves data, but its protection is spent. The rebuild — recomputing the lost disk’s contents onto a replacement — reads every remaining disk in full. On large modern disks that takes hours or days, during which the array is under maximum read load, exactly when a second failure or an unreadable sector is most likely to appear. In RAID 5, a second failure during rebuild loses the array. In RAID 6, two failures are survivable, which is why RAID 6 is the prudent choice for large disks: the rebuild window is long, and the second failure is not hypothetical.

A hot spare — a disk held idle in the array, ready to rebuild automatically — shortens the window between failure and rebuild start, but not the rebuild itself. It is worth having where the array must survive unattended, and it does not replace monitoring: a failed disk that nobody notices is a degraded array waiting for its second failure. Replace failed disks promptly, verify the rebuild completes, and keep the controller firmware current, because a rebuild that stalls is protection in name only.

RAID is not a backup. It protects against disk failure, and only disk failure. It does not protect against deletion, ransomware, controller failure, fire, flood, theft, or a mistaken command that writes garbage across the array — RAID will faithfully mirror the garbage. A backup is a separate copy, on separate hardware, ideally off-site, with a tested restore. Every RAID level needs one, including RAID 6, including RAID 10. The array keeps the service running; the backup brings the data back.

Choosing a level: the decision, by use case

The choice starts from the use, not from the level. A home media server that must survive one disk failure and values capacity takes RAID 5: one disk of protection, the rest usable, with a rebuild window to respect. A database that must survive a failure during a rebuild and values uptime over capacity takes RAID 6: two disks of protection, slower writes, and no single-rebuild gamble. A small array of two disks that must simply not lose data takes RAID 1: half the capacity, the simplest recovery, and a rebuild that copies rather than computes. A performance array where the data is a copy of something held elsewhere takes RAID 0, or no RAID at all, because the protection is bought with capacity and speed that the use does not need.

The number of disks constrains the choice before any preference does. RAID 1 needs two disks. RAID 5 needs three. RAID 6 needs four. RAID 10 needs four, in two mirrored pairs. An array that will grow should be planned for its final disk count, because changing levels later — migrating from RAID 5 to RAID 6, or adding disks to a RAID 10 — is possible on some controllers and impossible on others, and it is never free: it takes time, it stresses the disks, and it carries its own risk. Buy the controller and the enclosure for the level the array will end at, not the level it starts at.

Two questions close the decision. First, what is the backup? Every level needs one, and the level that is chosen should be the one whose rebuild window the backup makes survivable — the backup is what restores the array if the rebuild fails, and an array without a backup is a level of protection wrapped around nothing. Second, what is the recovery procedure, and has it been tested? A RAID level whose rebuild has never been run is a theory, and theories fail at 3 a.m. Build the array, pull a disk, rebuild it, and confirm the data. The level that survives the test is the level to keep.

Frequently asked questions