RAID 5 gives more usable capacity. RAID 10 gives faster writes and a far safer rebuild. RAID 5 spends one drive on parity and survives one failure. RAID 10 spends half the capacity on mirrors. The difference that decides most cases is the rebuild: RAID 5 must read every surviving drive completely, which is exactly when a second failure is most likely.
- RAID 5 costs four disk operations per write. RAID 10 costs two
- A RAID 5 rebuild reads every remaining drive, for hours
- RAID 10 rebuild reads one drive: the mirror partner
- RAID 10 survives two failures only if they are in different pairs
- Neither is a backup, and that is the expensive misunderstanding
On this page
The mechanismHow each one stores the data
The two RAID levels solve redundancy differently, and the mechanism explains every number in the comparison.
RAID 5 combines striping with parity. Data is striped across the drives, and for each stripe a parity block is computed and stored on one of them, rotating so no single drive holds all the parity.
If a drive dies, every missing block is reconstructed by combining the remaining data with the parity. One drive of storage is consumed by parity, and that is the entire capacity cost.
RAID 10 combines mirroring with striping. Drives are paired, each pair holds two identical copies of the data, and stripes run across the pairs. No parity is computed. A failed drive is covered by its mirror partner, which holds a complete copy.
That difference produces the write penalty, which is the number most people have never had explained.
Writing to RAID 5 costs four drive operations for one logical write: read the old data, read the old parity, write the new data, write the new parity. There is no way around it for a partial stripe write, and it is why RAID 5 write performance is poor for random data and perfectly adequate for sequential.
Writing to RAID 10 costs two operations, one to each side of the mirror, with no parity to compute. That is why the random write performance difference between the two RAID levels is a factor of two to three rather than a few percent.
Read performance favors RAID 10 slightly too, since either half of a mirror can serve the data, but the read gap is small and the write gap is the one that shows up in production.
The arithmetic worked out, for 4 TB drives, so the trade is a number rather than a formula.
| Drives | RAID 5 usable | RAID 6 usable | RAID 10 usable | What RAID 10 costs you |
|---|---|---|---|---|
| 4 | 12 TB | 8 TB | 8 TB | 4 TB |
| 6 | 20 TB | 16 TB | 12 TB | 8 TB |
| 8 | 28 TB | 24 TB | 16 TB | 12 TB |
| 12 | 44 TB | 40 TB | 24 TB | 20 TB |
The gap widens with every drive, which is why RAID 5 keeps being chosen for large arrays and why RAID 6 exists: at twelve drives, RAID 6 costs one more drive than RAID 5 and buys the second failure that RAID 10 would have cost twenty terabytes to provide.
Rebuild riskThe rebuild, which is the real argument
This is the section that changes minds, and it is about risk rather than speed.
When a drive fails in RAID 5, the array is degraded and still working. To rebuild onto a replacement, the controller must read every block of every remaining drive to recompute the lost data.
On an array of 8 TB drives that recovery takes many hours, often more than a day, at full read load on drives the same age as the one that just died.
Two things can go wrong in that window, and both are more likely than people assume.
A second drive fails. Drives bought together and worked identically tend to fail near each other, and a rebuild is the heaviest load the storage ever sees. A second failure during a RAID 5 recovery loses all the data.
An unrecoverable read error appears. Consumer drives are specified at around one unreadable sector per 10 to the 14 bits read. Reading 6 TB is roughly 5 times 10 to the 13 bits, so reading several such drives completely puts you where encountering one is a realistic outcome rather than a remote one.
During a normal read that is one bad file. During a RAID 5 recovery it can fail the whole rebuild.
RAID 10 does not have this problem in the same way. Recovery of a failed drive means copying its mirror partner: one drive read, no parity computation, no load on the rest of the storage. It is faster, and only one specific other drive can kill you during it.
That is the honest version of the argument. RAID 5 is not obsolete, and its fault tolerance runs out at exactly the period when you are already having a bad day.
Fault toleranceWhich failures each one survives
Stated precisely, because vendor summaries blur this.
RAID 5 fault tolerance is exactly one drive. Always one, whatever the array size. A second failure before recovery completes loses all the data.
RAID 10 fault tolerance is one drive always, and often two. With four drives in two mirrored pairs, losing one drive from each pair is survivable and losing both drives of one pair is not.
So a four drive RAID 10 survives two failures in about two thirds of the possible cases, and the marketing claim of two drive tolerance is a probability rather than a guarantee.
RAID 6 has two drives of fault tolerance always, at the cost of two drives of storage and a heavier write penalty of six operations. It exists specifically because RAID 5 recovery risk became unacceptable on large drives, and on arrays above about six drives it is usually the right comparison rather than RAID 5.
ChoosingChoosing, by workload
The decision is almost always determined by what writes to the array.
A database, a busy virtual machine datastore, or a mail server: RAID 10. These are random write workloads and the write penalty is the whole cost. The storage capacity you give up buys write performance you actually need.
A file server, an archive, or a backup target: RAID 5 or RAID 6. Large sequential writes, mostly reads afterward, and storage capacity is the point. Use RAID 6 rather than RAID 5 once the drives are large or the array is wide.
Anything with fewer than six large drives where losing the data is unacceptable: RAID 10. The recovery argument dominates at this size and the capacity difference is a few terabytes.
A boot pair or a hypervisor host: RAID 1, which is mirroring without the striping and is right for exactly two drives.
Very large arrays: neither, usually. Above ten or twelve drives the answer is RAID 6, or an erasure coded storage system that spreads the recovery load rather than concentrating it.
PitfallsWhere people go wrong
Treating RAID as a backup. It is not one, and this is the most expensive misunderstanding in storage. RAID fault tolerance covers a drive failing. It does not cover deletion, ransomware, a controller writing corruption to every drive, fire, or theft. Any array without a separate backup is one mistake from total data loss.
Building RAID 5 from large modern drives. The recovery window on 8, 12 or 16 TB drives is long enough that the risk stops being theoretical. Use RAID 6 or RAID 10 at those storage capacities.
Not having a hot spare. Recovery cannot start until a replacement drive exists. A hot spare starts it in seconds rather than whenever somebody notices the alert and finds a drive.
Buying every drive from the same batch. Identical drives with sequential serial numbers, run identically, fail at similar times. Mixing batches or vendors slightly reduces the chance of correlated failure during the exact recovery window where it matters most.
Not monitoring, or monitoring into a void. An array runs degraded indefinitely and quietly. Every real world catastrophe here involves an array that had been running on one failed disk for weeks. Alert to a person, and test that the alert arrives.
Assuming the RAID controller battery still works. Write back caching without a working battery or capacitor turns a power loss into data corruption across the array. That battery has a lifespan, and replacing it is scheduled maintenance rather than an event.
Expanding an array without a backup first. Reshaping is a long read and write across every drive, with exactly the same failure risk as a recovery, and it is voluntary. Back the data up before starting.
ComparisonRAID 5, RAID 6 and RAID 10, on capacity, writes and what recovery costs
| Criterion | RAID 5 | RAID 6 | RAID 10 |
|---|---|---|---|
| Minimum disks | Three | Four | Four |
| Usable capacity | N minus 1 | N minus 2 | Half |
| Failures always survived | One | Two | One |
| Write penalty | Four | Six | Two |
| Random write performance | Slow | Slowest | Fast |
| Rebuild reads | Every disk | Every disk | One disk |
| Rebuild risk on large disks | High | Lower | Lowest |
| Right for a database | No | No | Yes |
| Right for a large archive | Sometimes | Yes | Wasteful |
The rebuild row is the one to read twice. Everything else is a trade you can price. That row is a risk you carry.
FAQFrequently asked questions
What is the main difference between RAID 5 and RAID 10?
RAID 5 uses parity to give more usable capacity. RAID 10 uses mirroring to give faster writes and a much safer rebuild. Capacity against write speed and rebuild risk.
Which is faster, RAID 5 or RAID 10?
RAID 10, substantially, on random writes, because it costs two disk operations per write against four. Sequential reads are close.
How many disks can fail in RAID 10?
One always. Two if they are in different mirrored pairs, which on a four disk array is about two thirds of the possible combinations. Both disks of one pair loses the array.
What is the RAID 5 write penalty?
Four disk operations for every logical write: read the old data, read the old parity, write both back. It is inherent to parity and cannot be configured away.
Why is a RAID 5 rebuild risky?
It reads every block of every remaining disk, which is the heaviest load the array ever sees, on disks the same age as the one that failed. A second failure or an unrecoverable read error during that window loses everything.
Is RAID 5 obsolete?
No, and it is the wrong choice for large disks or write heavy workloads. For a small read heavy array where capacity matters, it still makes sense.
Should I use RAID 6 instead?
On arrays above about six disks, or with drives of 8 TB and up, usually yes. It survives two failures always, at one more disk of capacity and a heavier write penalty.
How much usable space does each give?
RAID 5 gives the capacity of all disks minus one. RAID 10 gives exactly half. On six 4 TB disks that is 20 TB against 12 TB.
Is RAID a backup?
No. It survives a disk failing and nothing else. Deletion, ransomware, corruption, fire and theft all defeat it completely.
What is an unrecoverable read error?
A sector the drive cannot read. Consumer drives are rated around one per 10 to the 14 bits, which is a realistic event when reading several multi terabyte disks completely, as a rebuild does.
Do I need a hot spare?
It is strongly worth having. The rebuild begins immediately rather than when somebody notices, and that shortens the window where the array has no redundancy.
Can I convert RAID 5 to RAID 10?
Some controllers can reshape in place and it is a long, heavy operation with the same risks as a rebuild. Back up first, and consider rebuilding the array fresh instead.
How long is RAID rebuild time?
RAID rebuild time depends on disk size, array load and the RAID level, and it runs from hours to days with large disks. A RAID 5 rebuild reads every remaining disk in full to recalculate the missing data, which is slow and stresses the array. A RAID 10 rebuild only copies one mirror, so it finishes sooner.
Keep readingRelated concepts
Read next · Disks and drives SSD vs HDD What goes in the array, and why rebuild times are a mechanical drive problem more than a flash one. Open this next12 min- Disks and drives · 10 min MBR vs GPT A RAID array presents one volume to the operating system, and that volume still needs partitioning.
- Shared storage · 12 min NAS vs SAN The array is inside something, and what that something serves decides the rest of the design.
- Backup · 13 min The 3-2-1 Backup Rule, and What Ransomware Did to It The redundancy that keeps a server running, and why it is not one of your three copies.
- Disks and drives · 9 min Btrfs vs ext4, and When Snapshots Are the Point Which Linux file system belongs on top of that array, and the RAID profiles Btrfs says to avoid.