How RAID 10 capacity works
RAID 10 — sometimes written RAID 1+0 — pairs drives into mirrors, then stripes data across those pairs. Every byte is written twice, so you give up exactly half the raw capacity. That number does not move with drive count.
Twelve 8 TB drives give you 48 TB usable from 96 TB raw. Four drives give you 16 TB usable from 32 TB raw. The 50% efficiency is a constant, which makes RAID 10 the easiest array to size and the hardest to justify on capacity alone.
Fault tolerance is not what the drives-per-pair number suggests
The common shorthand is "RAID 10 survives one drive failure." That is the guarantee, not the limit. The array survives any number of failures as long as no mirror pair loses both members. With eight drives you could lose four and keep running — if each failure lands in a different pair.
So the useful question is not "how many can fail" but "if one already failed, what happens on the next one?"
After one drive is gone, one remaining drive is its mirror partner — that one is fatal. Every other remaining drive is in a different pair and safe. Hence the formula.
| Drives | Mirror pairs | Usable | Survives 2nd failure |
|---|---|---|---|
| 4 | 2 | 50% | 67% |
| 6 | 3 | 50% | 80% |
| 8 | 4 | 50% | 86% |
| 12 | 6 | 50% | 91% |
| 16 | 8 | 50% | 93% |
This is the precise trade against RAID 6, which survives two failures with certainty. RAID 10 is probably two-failure tolerant and RAID 6 is certainly two-failure tolerant. On a four-drive array that difference is a coin flip weighted slightly in your favour. On a sixteen-drive array it is close enough not to matter much.
Why RAID 10 wins on writes
Parity RAID turns every small write into a read-modify-write cycle: read the old data, read the old parity, recompute, write both back. RAID 10 computes nothing. A write goes to both drives of the target pair and the operation is finished.
That gives RAID 10 a write penalty of 2, against 4 for RAID 5 and 6 for RAID 6. Reads can be served by either member of a pair, so read throughput scales with the full drive count.
| Level | Min drives | Usable | Fault tolerance | Random write |
|---|---|---|---|---|
| RAID 0 | 2 | 100% | None | Fastest |
| RAID 1 | 2 | 50% | 1+ (all copies) | Same as single drive |
| RAID 5 | 3 | (n−1)/n | 1 drive, certain | Slow — parity |
| RAID 6 | 4 | (n−2)/n | 2 drives, certain | Slowest — double parity |
| RAID 10 | 4 | 50% | 1 guaranteed, more if spread | Fast — no parity |
Rebuild time is the real risk, not the failure
A drive failing is an inconvenience. A rebuild that takes three days is an exposure window — the array is degraded the whole time, and every remaining drive is working harder than normal.
RAID 10 rebuilds are cheap because only the failed drive's mirror partner has to be read. Nothing has to be recomputed from parity across the whole array, which is why RAID 10 rebuilds in hours where RAID 5 or 6 on large drives can take days.
When RAID 10 is the wrong choice
Archive and backup targets. Sequential write workloads do not benefit from the absence of parity. RAID 6 gives you back a third of your capacity for a write penalty you will never notice on a nightly backup job.
Media libraries and file shares. Same reason. Streaming reads and large sequential writes are the workload parity RAID handles best.
Cost-sensitive arrays on large drives. At 20 TB per drive, the capacity you surrender to mirroring is expensive. Work out cost per usable TB before committing.
RAID 10 is the right answer for databases, virtualisation hosts, mail stores and anything with sustained random write traffic and a latency budget.
One thing this calculator will not fix
RAID is availability, not backup. A mirrored array survives a dead drive and does nothing at all against a deleted file, an encrypted volume, a controller that writes garbage to both mirrors, or a power event that corrupts the filesystem. If the array is your only copy, you do not have a backup — you have a more expensive single point of failure.
Questions
- Can I use drives of different sizes?
- Yes, but every pair is limited by the smallest drive in it. Mixed sizes waste capacity, and mixing speeds or cache sizes degrades the whole array to the slowest member. Match drives where possible.
- Does a hot spare change the usable capacity?
- Yes. A hot spare is not part of the array — it is idle until a failure triggers a rebuild. Subtract it from the drive count before calculating capacity. The trade is losing capacity now to shorten the degraded window later.
- Why does my 8 × 8 TB array show less than 32 TB usable?
- Two reasons. First, drive vendors sell decimal terabytes while operating systems report binary tebibytes, which accounts for roughly 9%. Second, filesystem metadata consumes a few percent more. Set the reserve field to match your filesystem.
- Is RAID 10 better than RAID 6 for a database?
- For write-heavy transactional databases, usually yes — the write penalty difference is large and predictable. For read-mostly analytical workloads on a budget, RAID 6 can be the better trade. The deciding factor is your write ratio, not a general rule.