RAID Levels Explained

From RAID 0 through RAID 1, 5 and 6 to compound levels — a complete overview

Architecture · Properties · Use-case recommendations · Hardware checklist · FAQ

On this page

RAID 0 — Data Striping NO REDUNDANCY  MAX. PERFORMANCE  MIN. 2 DISKS

Strictly speaking, RAID 0 is not a real RAID because it provides no redundancy. RAID 0 increases transfer rates by dividing the participating hard disks into contiguous blocks of equal size. These blocks are arranged like a zip fastener into one large virtual disk, so that accesses can be performed in parallel across all of them (striping means “splitting into stripes”, derived from stripe).

Throughput gains (in sequential access patterns, but especially under sufficiently high concurrency) come from the fact that the necessary disk operations can be carried out in parallel to a far greater extent. The block size is called the stripe granularity (also known as “chunk size” or “interlace size”). With RAID 0, a chunk size of 128 kB is most commonly chosen.

However, if one of the disks fails completely due to a defect, the RAID controller can no longer fully reconstruct the user data without that disk's contents. Partial restoration may be possible — namely for the files that happen to live entirely on the surviving disks — but this typically applies only to small files and only with a large stripe granularity.

RAID 1 — Mirroring FULL REDUNDANCY  MIN. 2 DISKS  50 % USABLE

A RAID 1 must consist of at least two hard disks holding exactly the same data (mirroring or duplexing). RAID 1 provides full redundancy of the stored data, while the array capacity is at most equal to the smallest participating disk. If one of the mirrored disks fails, any other can continue to serve all data. This is essential for safety-critical real-time applications. RAID 1 offers high fault tolerance: total data loss requires the failure of all disks. A typical use case is the boot volume in servers.

RAID 2 — Hamming

HISTORICAL · NO LONGER IN USE

RAID 2 is no longer used and is no longer supported by any manufacturer on the market. Its price/performance ratio is poor compared to RAID 5. The technique used 2 or more disks for data and one or more additional ECC disks. ECC stands for Error Correction Code. Each bit was distributed across all data disks, and an ECC value was generated and stored for it. This made it possible to verify data integrity on every access and correct errors when they occurred. Because the disks were addressed independently, extremely high transfer rates were possible — but only with synchronised disks. The effort required to achieve this is simply not worth it.

RAID 3 — Byte-Striping with Dedicated Parity Drive

PREDECESSOR OF RAID 5 · NO LONGER COMMON

RAID 3 is generally regarded as the predecessor of RAID 5. Striping means that the files to be stored are split across the available hard disks. Redundancy (loosely: the extent to which data is duplicated or reconstructible) is stored on a separate disk. The redundancy is calculated by adding the bits together and computing the sum. This allows the missing bit to be “back-calculated” when a disk fails. To illustrate: take 2 data disks and one disk that stores parity. The bits on the data disks are summed, and the result is either even or odd. That defines the sum bit on the parity disk:

If one of the two data disks fails, the missing value can be back-calculated:

Did it click? The elegant part: the same principle scales to arbitrarily many disks with just a single parity disk (limited only by write speed and capacity):

RAID 3 has since disappeared from the market, largely replaced by RAID 5, which distributes the parity evenly across all disks. The dedicated parity disk was a bottleneck. Before the transition to RAID 5, RAID 3 was partially improved by RAID 4, which standardised I/O operations on larger block sizes for performance reasons. It is also worth noting that a RAID 3 made of just two disks is by definition identical to a RAID 1 of two disks.

RAID 4

BLOCK-LEVEL PARITY · PARITY DISK BOTTLENECK

RAID 4 is similar to RAID 3 in that parity information is also calculated and written to a separate disk. The difference is that larger data blocks (stripes, chunks) are used instead of single bytes, which improves overall performance. RAID 5 works the same way. RAID 4 is often described as the “in-between” of 3 and 5. As with RAID 3, however, the parity disk remains the bottleneck despite the chunk-oriented approach.

RAID 5 — Performance + Parity

MOST POPULAR VARIANT  MIN. 3 DISKS  (N−1)/N USABLE

RAID 5 offers both increased read throughput and redundancy at relatively low cost, which makes it the most popular RAID variant. In write-intensive environments with small, non-contiguous changes, RAID 5 is not recommended, because the throughput drops significantly under random write loads due to the two-phase write procedure (a RAID 0+1 configuration would be preferable here).

RAID 5 is one of the most cost-efficient ways to store data redundantly across multiple hard disks while still using the available capacity efficiently. Because of the high controller requirements and prices, this advantage typically only kicks in beyond four disks. The user data is distributed across all disks just as in RAID 0. The parity information, however, is not concentrated on one disk as in RAID 4 but distributed as well. Suitable RAID controllers with cache and BBU/flash backup are recommended for production environments.

RAID 6 / DP — Redundancy via Two Additional Disks

TOLERATES 2-DISK FAILURE  MIN. 4 DISKS  DOUBLE PARITY

RAID 6 (Advanced Data Guarding) works similarly to RAID 5 but tolerates the failure of up to two disks. Two (rather than one) error-correction values are calculated and distributed across the disks in such a way that data and parities lie block-by-block on different disks. This means a gross capacity of n+2 disks for a net data capacity of n disks, which still represents a cost saving over simple mirroring (RAID 1) even with relatively small net disk counts.

The computational overhead of the underlying XOR operations, however, is considerably higher than with RAID 5. RAID 5 calculates a parity bit by adding the data bits in a single row (and resynchronises by adding the row data back). RAID 6 must compute the parity across multiple rows — resynchronisation, especially with two failed disks, requires calculations involving matrices and inverse matrices from linear algebra (coding theory). A RAID 6 array requires at least four disks!

RAID TP

TOLERATES 3-DISK FAILURE  PROPRIETARY EASYRAID

RAID TP is a proprietary RAID with triple parity from the manufacturer easyRAID. According to the manufacturer, RAID TP tolerates the failure of up to three disks. Data blocks and parities are written in parallel to the individual physical disks. The three parities are placed on different stripes on different disks. The RAID Triple Parity algorithm uses a special code with a Hamming distance of at least 4. This requires a minimum of four disks. The capacity is calculated as the number of disks minus three.

RAID 7

EXPENSIVE · RARELY USED

Same as RAID 5, but the RAID controller runs a real-time operating system and faster data buses are used, which leads to significantly higher transfer rates. RAID 7 is rarely deployed because it is very expensive.

RAIDnm

FREELY CONFIGURABLE FAULT TOLERANCE

RAIDn extends the principle of generating multiple parity values from RAID 6. This can be used to absorb an arbitrarily large number of disk failures. Such arrays are referred to as RAID nm, RAID (n,m) or RAID n+m. The factor n describes the total number of disks used and the factor m the number of disks that may fail.

JBOD — Just a Bunch Of Disks

NOT A RAID · NO REDUNDANCY

“JBOD” stands for Just a Bunch Of Disks. The term is used in contrast to RAID systems to indicate that disks are operated individually rather than as a unified array. Many hardware RAID controllers can present the connected disks individually to the operating system; the controller's RAID functions are disabled and it operates as a simple disk controller.

A JBOD can also — independently of the controller — refer to any group of disks attached to the computer. With the help of volume management software, such a JBOD can be aggregated into a software RAID. In some RAID tools, such as the VIA RAID-TOOL, the term JBOD is sometimes (incorrectly) used synonymously with NRAID.

RAID 10 or RAID 0+1

FAST WRITES  MIN. 4 DISKS  50 % USABLE

This RAID level is a combination of RAID 1 (mirroring) and RAID 0 (striping) and combines the benefits of both — security and sequential performance. It is sometimes also called RAID 0+1. It usually consists of 4 disks, since RAID 10 is built from two pairs of mirrored arrays that are then combined into a RAID 0 array. RAID 10 is particularly suitable for large files that need to be stored redundantly; because no parity has to be calculated, write operations are very fast — ideal for databases and VM hosts on rack-mount servers equipped with NVMe / M.2 SSDs or SAS SSDs.

RAID Matrix

FLEXIBLE · MIXED OPERATION  INTEL PROPRIETARY

This Intel-developed technology is a low-cost and flexible RAID solution. Matrix RAID allows different RAID types to be mixed on a single set of drives. Part of the drives can be configured as RAID 1 (full mirroring) while the performance-sensitive part runs as RAID 0. In practice this means: the operating system partition including critical data is protected by a RAID 1, while performance-hungry applications run on a separate RAID 0 partition at maximum speed.

RAID 50

DATA SECURITY + PERFORMANCE  MIN. 6 DISKS

Developed by AMI, RAID 50 is recommended when data security, fast access times and high transfer rates are required at the same time. RAID 50 can be implemented with a minimum of 6 disks. Typical use cases are storage servers and larger direct-attached storage solutions.

Consulting on RAID configuration and storage hardware

Planning a new server, a storage extension or upgrading an existing array? Our team supports you in selecting the right RAID controllers, hard disks, SSDs and matching drive caddies — for standard servers as well as for SAN and NAS systems.

Phone: +49 (0)7666 / 88499-0  ·   E-mail: sales@industry-electronics.com

Further reading

RAID Systems — hardware architectures and selection

A conceptual introduction to RAID systems — history, benefits, PC-based vs. platform-independent disk arrays, selection criteria for controllers and external storage subsystems.
» To the article: RAID Systems — Hardware Architectures, Benefits, Selection

Backup Strategies — why RAID is not a backup

A RAID increases availability against hardware failure, but is not protection against ransomware, accidental deletion or logical errors. Our Backup Strategies series covers risks, RTO/RPO, the 3-2-1 rule and concrete methods, media and software.
» To Part 1: Backup Fundamentals — risks, RTO/RPO, 3-2-1 rule