US What RAID is
RAID, short for redundant array of independent disks, is a way of combining several drives into one logical volume. The computer sees a single drive, while a RAID controller or the operating system spreads the data across the members of the array. A RAID array can be faster than one drive, keep working when a drive fails, or both.
The idea comes from a 1988 University of California, Berkeley paper by Patterson, Gibson and Katz, which used “inexpensive” in place of “independent”. The numbered RAID levels it introduced are still how arrays are described today, and the Storage Networking Industry Association (SNIA) maintains the common definitions.
How RAID works: striping, mirroring and parity
Every RAID level is built from three techniques, used alone or together.
-
//
Striping. Data is split into chunks written across several drives in turn, so reads and writes are shared between them. It adds speed but no protection.
-
//
Mirroring. Every write goes to two or more drives, so each holds a complete copy. If one fails, the other carries on.
-
//
Parity. The array stores calculated check data alongside the data. If a drive fails, its contents can be rebuilt from the parity and the surviving drives.
-
//
Hot spares. An extra drive kept idle in the array. When a member fails, the controller starts rebuilding onto the spare straight away.
RAID levels compared
Five RAID levels cover almost every real system. The table assumes drives of equal size, written as n drives of capacity C.
| RAID level | Minimum drives | Usable capacity | Drive failures survived | Best suited to |
|---|---|---|---|---|
| RAID 0 (stripe) | 2 | n × C | None | Scratch space and temporary data that can be recreated |
| RAID 1 (mirror) | 2 | C (one drive’s worth) | All but one drive | Boot and operating system drives |
| RAID 5 (striping with parity) | 3 | (n − 1) × C | One | Read-heavy file storage on smaller drives |
| RAID 6 (double parity) | 4 | (n − 2) × C | Any two | Large arrays and high-capacity hard drives |
| RAID 10 (mirrored stripes) | 4 | n × C ÷ 2 | One per mirror pair | Databases, virtual machines and write-heavy recording |
Other levels exist. RAID 2, 3 and 4 are rarely used today, and RAID 50 and 60 stripe across several RAID 5 or RAID 6 groups in large arrays.
RAID 5 vs RAID 6 and RAID 10
Parity costs write performance. Each small random write to RAID 5 takes four drive operations (read data, read parity, write data, write parity), RAID 6 takes six, and RAID 10 only two. This write penalty is why a parity array that looks fast for large files can struggle with databases, and why IOPS should be calculated for the array, not the drive.
Rebuild time is the other factor. When a drive fails, every surviving drive is read end to end to rebuild it. On multi-terabyte hard drives that can take a day or more, and a second failure, or an unreadable sector, during that window loses a RAID 5 array. RAID 6 survives it, which is why it has replaced RAID 5 for large hard drive arrays. RAID 10 rebuilds fastest, because it only copies one mirror.
RAID 10, sometimes written RAID 1+0, is worth understanding on its own. Drives are first paired as mirrors, then data is striped across the pairs. A four-drive RAID 10 gives half the raw capacity, reads and writes faster than RAID 5 or 6, and survives one failure in each pair. The cost is capacity: you always lose half.
Hardware, software and firmware RAID
RAID can be run in three places, and the choice affects performance, recovery and portability.
-
//
Hardware RAID. A dedicated controller card with its own processor and cache, often protected by a battery or supercapacitor. It suits SAS and SATA hard drive arrays and is independent of the operating system.
-
//
Firmware or chipset RAID. Built into the motherboard and set up in the UEFI menus, such as Intel Rapid Storage Technology or Intel VROC for NVMe SSDs. Common on server motherboards and many industrial boards.
-
//
Software RAID. Managed by the operating system, such as Linux mdadm, Windows Storage Spaces or ZFS. It needs no extra hardware and moves easily between machines.
RAID in industrial and edge systems
At the edge, RAID is mostly about keeping a system running until someone can reach it. Mirrored boot drives keep a remote computer booting after a drive failure, and parity or RAID 10 arrays protect recorded data. Many of our industrial NVRs support RAID across their drive bays for exactly that reason, so a failed drive does not interrupt recording.
In vehicles and mobile systems, the drives are usually SSDs in removable or hot-swap trays, and many rugged computers pair two or more of these bays with RAID 1 so a drive can be replaced at the depot without losing the operating system. Whatever the level, pair RAID with monitoring that raises an alert when a drive fails, because a degraded array that nobody notices offers no protection at all.
Quick answers on RAID
RAID: common questions
What does RAID stand for?
Redundant array of independent disks. The original 1988 paper said “inexpensive disks”; the industry later changed it to “independent”.
What is RAID 0?
Striping with no redundancy. Data is split across two or more drives for speed and full capacity, but if any drive fails, the whole array is lost.
What is RAID 1?
Mirroring. Two drives hold identical copies, so the system keeps running if one fails. You get the capacity of one drive.
Is RAID a backup?
No. RAID survives drive failure, but deletions, ransomware and corruption are copied to every drive straight away. Keep separate backups, ideally off site.
Which RAID level is best?
It depends on what you are protecting. RAID 1 for a boot drive, RAID 10 for databases and heavy writes, RAID 6 for large hard drive arrays. RAID 0 only where the data can be lost.
Can you use SSDs in RAID?
Yes. SSDs work in any RAID level. RAID 1 and RAID 10 are common for NVMe SSDs, using chipset RAID such as Intel VROC or software RAID.
We set up the RAID level, drives and monitoring for every multi-drive system we supply as standard, then validate the array before the system ships.