RAID levels come up in nearly every storage decision we write about, from PERC controllers to backup servers, and they're usually mentioned as though everyone already knows what RAID 6 means and why it beats RAID 5 on large drives. Most buyers don't, and the guides that do explain it tend to be written for home labs or for vendors.
This is the business version: what each level actually does, what it costs in capacity and write performance, why the rebuild window matters more than the fault-tolerance number, and which level belongs under which workload on a Dell or HPE server.
RAID 10 for anything that writes randomly. RAID 6 for anything that stores in bulk.
RAID 1 for the boot drives, and no RAID 5 on drives above 4 TB. RAID protects against drive failure — not deletion, ransomware, corruption or fire. RAID is availability; backup is recovery. You need both.
- The workload is random and write-heavy: databases, VM datastores, busy app servers
- You want fast rebuilds that don't stress the array
- You're on SSDs, where speed matters more than the 50% capacity cost
- The workload is sequential or read-heavy: file shares, backup, archive, CCTV
- You're on large hard drives, where capacity per riyal is the point
- You want to lose only two drives to parity across eight or twelve
Three building blocks
Every RAID level is a combination of three ideas. Understand these and the numbered levels stop being arbitrary.
| Technique | What it does | Gives you | Costs you |
|---|---|---|---|
| Striping | Splits data across drives so several work at once | Speed and capacity | No protection at all |
| Mirroring | Writes the same data to two drives | Simple, fast protection and rebuilds | Half your capacity |
| Parity | Stores a calculated checksum that can rebuild a missing drive | Protection for the cost of one or two drives | Write performance and slow rebuilds |
The levels, one by one
RAID 0
Striping, no protection. Minimum 2 drives.
- Full capacity, fastest reads and writes
- Lose any one drive and you lose everything
- Only for scratch space that can be regenerated, like a render cache
RAID 1
Two drives holding identical copies. The standard for boot volumes, like a Dell BOSS card.
- Carries on without interruption if one drive fails
- Rebuild is a straight copy, not a calculation
- Usable capacity is one drive
RAID 5
Striping with single parity. Minimum 3 drives.
- All drives minus one usable
- Fast reads
- Slow writes: four disk operations per write
- No protection left during a long rebuild on large drives
RAID 6
Striping with double parity. Minimum 4 drives. What makes large hard-drive arrays safe.
- Survives any two drive failures
- Stays protected while it rebuilds
- All drives minus two usable; reads unaffected
- Slower writes: six disk operations per write
RAID 10
Mirrored pairs, striped together. Minimum 4 drives. The level for databases and VMs.
- Best random-write performance of any level
- No parity maths; rebuilds by simple copy
- Half the capacity
- A second failure in the same mirror pair is fatal
RAID 50 and 60
Several RAID 5 or RAID 6 groups, striped together. For large arrays, typically 16 drives and up.
- Faster rebuilds than one big parity group
- RAID 50 survives one failure per group; RAID 60, two
- Each group gives up its own parity drives
Splitting twelve drives into two RAID 6 groups of six, then striping across them, rebuilds faster and tolerates more failures than one twelve-drive RAID 6 — that's the whole case for RAID 60.
Side by side
| Level | Min drives | Survives | Usable capacity | Write speed | Rebuild |
|---|---|---|---|---|---|
| RAID 0 | 2 | Nothing | 100% | Fastest | None possible |
| RAID 1 | 2 | 1 drive | 50% | Fast | Fast, simple copy |
| RAID 5 | 3 | 1 drive | (N−1)/N | Slow, 4 IOs per write | Slow, risky on large drives |
| RAID 6 | 4 | 2 drives | (N−2)/N | Slower, 6 IOs per write | Slow, but protected during it |
| RAID 10 | 4 | 1 per mirror pair | 50% | Fast, 2 IOs per write | Fast, simple copy |
| RAID 50 | 6 | 1 per group | Depends on groups | Moderate | Faster than one large RAID 5 |
| RAID 60 | 8 | 2 per group | Depends on groups | Moderate | Faster than one large RAID 6 |
The write penalty, and why it matters for databases
The write-speed column is the one buyers underestimate. On a mirrored level, one logical write becomes two physical writes. On RAID 5 it becomes four: read the old data, read the old parity, write the new data, write the new parity. On RAID 6 it's six, because there are two parity blocks to update.
For sequential writes, such as a backup job streaming to disk, the controller can often write whole stripes at once and the penalty largely disappears. For small random writes — exactly what a transactional database or a busy VM host generates — the penalty is real and constant. That's why RAID 10 is the standard for databases and VMs, and RAID 5 or 6 for file storage and backup targets.
A dead RAID battery silently makes RAID 5 slower. A controller with battery-backed write cache, such as the PERC H730P or H740P, hides much of the parity penalty by acknowledging writes to cache. When the battery fails, the controller drops to write-through mode and the full penalty comes back. Replacement batteries, like the HPE 96W Smart Storage Battery, are in stock.
The rebuild window: why RAID 5 stopped being safe
"Survives one drive" is true only at the instant the drive fails. What matters is what the array has to do next: read every remaining drive end to end and rebuild the lost one. During that window, a RAID 5 array has no protection left. It's a RAID 0.
-
Drives got bigger, not fasterA 2 TB drive rebuilds in a few hours. A 12 or 16 TB drive under production load can take a day or more — and drives from the same batch, same age, same environment are the ones most likely to follow the first.Longer window
-
Every sector has to read correctlyA rebuild reads every sector of every surviving drive. On a large array of large drives, the chance of hitting one unreadable sector becomes real — and on RAID 5, one unreadable sector fails the rebuild.More risk
Enterprise drives are rated an order of magnitude better than consumer ones, and real-world error rates are lower than the spec sheet suggests, so it's less dire than the worst calculations imply. But the conclusion holds: the rebuild window is long, the array is unprotected during it, and a second problem of any kind is fatal.
RAID 6 is the floor for any hard-drive array on drives above about 4 TB — and for any array whose job is to be the copy you fall back on.
With two parity blocks, the array stays protected against a second failure or an unreadable sector for the entire rebuild. That's the whole argument.
RAID 10 or RAID 6: the real business decision
In practice, most business server storage comes down to these two. RAID 0 is out, RAID 1 is for boot drives, and RAID 5 has been superseded. The choice is performance versus capacity — see the short answer above for which fits where.
Many servers reasonably run both: a RAID 10 set of SSDs for the VMs and databases, and a RAID 6 set of hard drives for file shares and local backup. Modern PERC and Smart Array controllers run several virtual disks on one controller without difficulty.
Hot spares, cold spares and patrol reads
Hot spare
A drive installed, powered and idle. The controller rebuilds onto it the moment a drive fails — the exposed window starts immediately instead of whenever someone notices. On any array that matters, configure one.
Cold spare
A matching drive on the shelf. Keep one even with a hot spare: once the hot spare is used you're waiting for a delivery, and in Qatar that can take weeks.
Patrol read
Patrol read and consistency checks scan the array in the background, remapping unreadable sectors before a rebuild depends on them. Leave them enabled.
Alerting
A degraded array nobody knows about is how a survivable failure becomes data loss. Set iDRAC or iLO to email on drive events — and test that it works.
Hardware RAID or software RAID?
Everything above applies to both; the difference is where the work happens. A hardware controller (PERC, Smart Array) does the parity maths and caching in dedicated silicon, presents the array as one disk, and is what Windows Server, VMware and Veeam expect. Software RAID — ZFS in particular — does the work on the host CPU, needs raw drive access through an HBA, and adds checksumming and self-healing that hardware RAID can't.
The right choice depends on the platform on top, covered in our Ceph, ZFS and hardware RAID comparison.
Never layer software RAID on top of a hardware array. Pick one.
Which level for which workload
-
Boot / OS volumeTwo SSDs or a BOSS card. Keep the OS off the data array.RAID 1
-
Virtual machine datastoreRandom and write-heavy. SSDs preferred.RAID 10
-
Transactional databaseNever parity RAID under an OLTP database.RAID 10
-
File server, document sharesRead-heavy, capacity matters, large drives.RAID 6
-
Backup targetDual parity without exception; RAID 60 above sixteen drives.RAID 6 / 60
-
CCTV / surveillance recordingSequential writes, large capacity. Our CCTV storage guide covers sizing.RAID 6
-
Archive / media libraryWrite once, read occasionally.RAID 6
-
Scratch / temporary render spaceOnly if the data can be regenerated from elsewhere.RAID 0
-
Any array on drives above 4 TBNot RAID 5 — the rebuild window is too long.RAID 6 · 10 · 60
We configure RAID on every refurbished Dell PowerEdge and HPE ProLiant we supply — controller, cache battery and drive layout matched to the workload you describe, with spare drives in stock in Doha.
Conclusion
The RAID decision for most business servers reduces to two choices: RAID 10 for anything that writes randomly, RAID 6 for anything that stores in bulk. RAID 1 for the boot drives. RAID 5 belongs to an era of small drives, and RAID 0 belongs nowhere important.
Then the three habits that make any level work: a hot spare in the chassis, a cold spare on the shelf, and alerting that actually reaches someone. An array that survives a failure nobody hears about is one more failure away from not surviving.
Common questions
Is RAID 5 still safe to use?
On small, fast drives with a hot spare and good monitoring, it's tolerable. On large hard drives it isn't recommended: the rebuild after a failure takes a day or more, the array has no protection during that window, and a second failure or unreadable sector loses the data. RAID 6 costs one extra drive and removes that risk.
RAID 10 or RAID 6 for a virtual machine host?
RAID 10. Virtual machines generate random writes, and parity RAID's write penalty hurts that workload directly. RAID 10 has no parity calculation, rebuilds quickly by simple copy, and delivers the best random-write performance of any level. Use RAID 6 for the file shares and backups, not the VMs.
Does RAID replace backup?
No. RAID keeps a server running when a drive fails. It does nothing against deletion, ransomware, corruption, controller failure, fire or theft — it faithfully replicates all of those. RAID is availability. Backup is recovery. Every array needs a backup behind it.
What is the RAID write penalty?
The number of physical disk operations one logical write generates. Mirroring is two. RAID 5 is four: read old data, read old parity, write new data, write new parity. RAID 6 is six. For sequential writes the controller can often avoid it; for small random writes it's constant, which is why parity RAID is a poor fit for databases.
How many drives should be in a RAID 6 array?
Six to twelve is the practical range. Below six, the two-drive parity cost is a large fraction of capacity. Above about sixteen, rebuild times and the read load during rebuild become unwieldy, and splitting into RAID 60 groups is better.
Where can I buy RAID-configured servers in Qatar?
ServerDove supplies refurbished Dell PowerEdge and HPE ProLiant servers across Qatar with RAID configured to your workload, cache batteries verified, and spare drives stocked in Doha for same-week replacement.
