RAID Levels Explained: 0, 1, 5, 6 and 10 for Server Storage

RAID level is usually chosen once, at build time, and then lived with for years. It determines how much of your raw capacity you actually get, how long a rebuild takes, and whether a second drive failure destroys the array or is survivable.

This guide covers the levels that matter in practice, what each costs you, and how drive capacity changes the answer.

What RAID does and does not do

Worth stating first because it causes real data loss. RAID is availability, not backup.

RAID protects against a drive failing. It does not protect against deletion, corruption, ransomware, a controller failure that scrambles the array, fire, theft or someone rebuilding the wrong volume. All of those propagate across every member instantly.

An array with no tested backup behind it is one mistake away from total loss regardless of RAID level. That is not an argument against RAID, it is a reminder that the two solve different problems.

RAID 0: no redundancy

Data is striped across drives with no parity and no mirror. You get the full capacity of every drive and the combined throughput of all of them.

Any single drive failure destroys the entire array. With more drives the risk multiplies rather than divides.

Legitimate uses are narrow: scratch space, video editing caches, render temporary storage — anywhere the data is reproducible and speed matters more than keeping it. It is never appropriate for anything you would miss.

RAID 1: mirroring

Two drives holding identical data. Usable capacity is half the raw total, which sounds expensive until you look at what it buys.

Rebuild is a straight copy. When a mirror member fails, the replacement is populated by copying from the survivor. No parity calculation, no reading every drive in the array. It is the fastest and least stressful rebuild of any RAID level.

That makes RAID 1 the standard choice for server boot volumes, where you need the OS to survive a drive failure and capacity is irrelevant.

One practical warning: when one member of an old mirror fails, the survivor has identical accumulated hours and often the same manufacturing batch. It is not a fresh drive. Many sites replace both.

RAID 5: single parity

Data and parity striped across all drives. One drive’s worth of capacity goes to parity, so a five-drive array gives you four drives of usable space. The array survives one drive failure.

The capacity efficiency is why RAID 5 was the default for years. The problem is what happens during a rebuild.

When a drive fails, the array reconstructs the missing data by reading every remaining drive in full. That takes hours or days depending on capacity, and throughout it the array has no redundancy left while every surviving drive is under sustained read load. Sustained load on drives of identical age is precisely the condition that provokes a second failure — and a second failure during a RAID 5 rebuild means total loss.

This is why RAID 5 is widely considered inadvisable with large drives. At 2TB, rebuilds run in hours and the exposure is short. At 12TB or 14TB, a rebuild can exceed a day and the risk becomes serious.

RAID 6: double parity

Same idea as RAID 5 with two drives’ worth of parity instead of one. A six-drive array gives you four drives of usable space, and it survives two simultaneous failures.

That second parity drive is doing one specific job: covering the rebuild window. If a second drive fails while the first is being reconstructed, RAID 6 keeps going where RAID 5 would be gone.

The costs are one more drive of capacity and slightly slower writes, since two parity calculations happen instead of one. On modern controllers the write penalty is rarely the limiting factor.

For any array built on drives of 4TB and above, RAID 6 or an equivalent double-parity scheme should be the default rather than the upgrade.

RAID 10: mirrored stripes

Pairs of mirrored drives, striped together. Usable capacity is half the raw total — the same overhead as RAID 1, applied across a larger array.

What you get for that is the best rebuild behaviour available. Because each drive has a mirror partner, a rebuild copies from one drive rather than reading the whole array. Rebuilds are fast, the array stays responsive throughout, and the load on surviving drives is minimal.

It also handles random write workloads better than parity levels, because there is no parity to calculate on every write.

That combination makes RAID 10 the common choice for virtualisation datastores and busy databases, where both write performance and rebuild behaviour matter. The capacity cost is high and frequently worth it. See our 1.2TB 10K guide for the drive class that usually sits under it.

Calculating what you actually get

Raw capacity is not usable capacity, and the gap surprises people who order drives against a target number.

Take twenty-four 1.2TB drives — 28.8TB raw.

RAID 10: roughly half, so about 14.4TB.
RAID 6: two drives to parity, so 22 drives of usable space, about 26.4TB.
RAID 5: one drive to parity, about 27.6TB — and not recommended at this scale.

Then subtract hot spares, which come off the top before any of this. Two hot spares leaves twenty-two drives to work with.

Decide RAID level and spare allocation before ordering drives, not after. Buying exactly enough drives to hit a raw number and then discovering the usable volume is far smaller is a common and avoidable problem.

Group width: fewer big groups or more small ones

A decision that gets skipped and matters more than it looks.

A very wide RAID group — twenty-four drives in one parity set — is capacity-efficient because parity overhead is spread thin. It also means a rebuild reads twenty-three drives, takes longer, and puts more drives at risk during the window.

Several smaller groups cost more in parity overhead and recover faster, with fewer drives exposed per rebuild. A failure affects one group rather than everything.

For bulk storage on large drives, several smaller groups is usually the better trade. For a modest array, one group is fine.

Hot spares and cold spares

Both, and they do different jobs.

A hot spare is a drive already installed and configured in the array, sitting idle. When a member fails the controller brings it in automatically and the rebuild starts immediately — not when someone notices the alert.

A cold spare is a physical drive on the shelf. Once the hot spare has been consumed, it restores full protection without waiting for delivery.

On platforms where the manufacturer no longer supplies parts — IBM System x, HP ProLiant G-series, NetApp FC shelves — the cold spare matters more than usual, because market availability rather than a supplier commitment is your recovery path. Our compatibility guide covers which part numbers fit which platform.

Rules that apply to every level

The group uses the smallest member capacity. A larger drive contributes only as much as the smallest, and the excess is stranded.

The group degrades to the slowest member. One 7200 RPM drive among 15K members slows everything. This is why dropping a single SSD into a mechanical array achieves nothing.

Tiering happens between groups, not within them. Fast and bulk storage in one chassis means separate groups from separate drive types.

Manufacturer does not need to match, provided capacity, interface, form factor and speed do. Mixing batches deliberately even helps, since drives from one batch tend to fail at similar times.

Choosing quickly

Boot volume: RAID 1. Virtualisation or busy database: RAID 10. Bulk storage on 4TB+ drives: RAID 6, in several groups rather than one wide one. Small array on drives under 2TB where capacity is tight: RAID 5 is defensible. Scratch space you can regenerate: RAID 0.

And whichever you choose, verify that your backups actually restore. RAID is availability.

Common questions

Tell us your bay count, drive capacity and workload and we will recommend a RAID level and spare allocation alongside the quote.