Wesloo
All topicsTopics

Data storage

RAID Levels Explained: Which RAID Level Should You Use?

Short answer

RAID is a method of combining several drives into one storage system to improve speed, capacity, or fault tolerance. Different RAID levels make different trade-offs among those goals, and none of them replaces a backup.

Key points

  • RAID 0 stripes data for speed but has no redundancy.
  • RAID 1 mirrors data and can survive one drive failure.
  • RAID 5 and RAID 6 use parity to recover from one or two failed drives.
  • RAID 10 combines mirroring and striping for strong performance and redundancy.
  • RAID protects against drive failure, not deletion, corruption, or ransomware.

What are RAID levels?

RAID levels are configurations that combine multiple hard drives or SSDs into one storage system. The goal can be higher throughput, larger usable capacity, protection against a failed drive, or a mix of these. Data may be striped across drives, mirrored to a second drive, or protected with parity, which is a calculated value used to rebuild missing data.

RAID is not a single standard. No standards body fixed the RAID level numbers, so vendors sometimes use their own numbers or implementations. That is why a NAS or server manual may list levels that do not match a textbook list. JBOD, short for just a bunch of disks, is not a RAID level; it presents drives separately and is often used for swap or spooling.

How RAID stores data

Striping splits data into blocks and writes them across several drives at the same time. A RAID 0 array with n drives acts like one large disk whose data rate is n times a single drive's rate, but the stripe size set when the array is created must stay fixed. Mirroring keeps an exact copy of data on another drive. Parity stores a calculated value that allows recovery when a drive fails.

RAID can also detect problems during data scrubbing, a process that reads stored data and checks it against redundant information. If a drive fails, RAID uses the mirror or parity data to reconstruct what was lost, although rebuilds can reduce performance on some levels.

RAID levels compared

LevelMinimum drivesFault toleranceUsable capacityBest for
RAID 02None100% of raw capacityTemporary high-speed data
RAID 12One drive50% of raw capacitySmall critical datasets
RAID 53One drive(n-1)/n of raw capacityGeneral storage
RAID 64Two drives(n-2)/n of raw capacityLarge arrays
RAID 104One drive per mirrored pair50% of raw capacityDatabases and virtual machines

Capacity is expressed as a fraction of the sum of all drive capacities. For example, three 250 GB drives in a parity array with one parity drive give 750 GB total and 500 GB usable. The same logic applies to RAID 5 with three drives.

RAID 0: speed without redundancy

RAID 0 stripes blocks across every drive in the array. It uses all capacity and delivers high throughput, but a single disk failure destroys the whole array because every file's blocks are spread across all drives.

Use RAID 0 only for data you can recreate or replace, such as render caches, temporary scratch space, or test datasets. It is a poor choice for anything you would hate to lose.

RAID 1: simple mirroring

RAID 1 keeps an exact copy of data on a second drive. It can survive one drive failure and allows immediate data access during that failure. Read performance can improve because either mirrored disk can serve a request, while write performance may not improve much because both copies must be updated.

RAID 1 is easy to understand and rebuild, but half of the raw capacity is spent on the mirror. It suits boot volumes, small business file shares, and systems where simplicity matters more than capacity.

RAID 5: parity with one drive's worth of capacity

RAID 5 stripes data at block level and stores parity across the array rather than on one dedicated disk. It can tolerate one disk failure. Usable capacity is one drive less than the total, so a four-drive array gives the capacity of three drives.

RAID 5 provides good random read performance. Writes carry parity overhead, so write speeds are lower than in RAID 0 or RAID 1. During a rebuild, performance may degrade, but data remains available.

RAID 4 also uses block-level striping, but it keeps parity on a dedicated disk. That layout gives good random read performance, while random write performance is low because every parity update goes to the same disk unless the filesystem is RAID-4-aware. RAID 3 is similar, but parity is always written to the same drive. These levels exist, but they are less common than RAID 5 and RAID 6.

RAID 6: surviving two failures

RAID 6 is like RAID 5 but stores two independent parity blocks. It can handle two simultaneous disk failures, which matters in large arrays where rebuilds take a long time and a second failure becomes more likely. The cost is two drives' worth of capacity and more parity calculation overhead.

Choose RAID 6 when capacity is large enough that rebuild time and second-failure risk outweigh the extra parity cost. It is common in NAS and server storage with many drives.

RAID 10: mirroring plus striping

RAID 10, also written RAID 1+0, creates mirrored pairs and then stripes across them. It combines the rebuild simplicity and read performance of mirroring with the throughput of striping. It needs at least four drives and yields half of raw capacity.

RAID 10 is often the best choice for databases, virtual machine stores, and transactional workloads that need low latency and predictable performance. Its main disadvantage is cost, because half of the purchased capacity is used for redundancy.

Nested RAID levels

Administrators sometimes combine standard levels. Common names include RAID 0+1, RAID 1+0 (RAID 10), RAID 5+0 (RAID 50), RAID 6+0 (RAID 60), and RAID 10+0 (RAID 100). These designs trade more drives and complexity for larger capacity or higher throughput while keeping some redundancy. They are useful in enterprise storage, but they also multiply rebuild and management complexity.

Software, firmware, and hardware RAID

The RAID logic can run in three places. Software RAID uses the computer's processor and memory. It needs no special hardware and is cost-effective, but it can reduce overall system performance and is slower than hardware RAID. Firmware RAID, also called fake RAID or hybrid RAID, is built into the BIOS or firmware and needs a driver after the operating system loads; it costs less than hardware RAID but still draws on CPU resources. A hardware RAID controller is a separate card and offers better performance, especially with RAID 5 and RAID 6, at a higher cost.

Some operating systems include software RAID. Windows Server 2012 and Mac OS X are examples. For a small system, software RAID may be enough; for a busy server, a hardware controller or a well-tested software stack with enough CPU and memory is usually better.

How to choose a RAID level

Start with the workload, not the level name. Sequential video or backup streams behave differently from random database I/O. Then decide how many drive failures you can tolerate, how much capacity you can afford to lose, and how long a rebuild can take.

  • Choose RAID 0 only for disposable, high-speed data.
  • Choose RAID 1 for small datasets where simplicity and quick recovery matter.
  • Choose RAID 5 for general storage when you have at least three drives and can accept one failure.
  • Choose RAID 6 for larger arrays where two-drive failure protection is worth two parity drives.
  • Choose RAID 10 for latency-sensitive or write-heavy workloads when budget allows.

If the array is large, remember that rebuilds stress every remaining drive. A second failure during rebuild can destroy a RAID 5 array, which is why RAID 6 or RAID 10 is often safer for big sets.

RAID is not a backup

Every RAID level except RAID 0 can survive one drive failure, and RAID 6 can survive two disks failing at once. Even so, RAID does not protect against accidental deletion, file corruption, ransomware, or a controller failure. For complete security, keep backups of the data on a RAID system too. A practical method is the 3-2-1 backup rule, which keeps multiple copies in separate places.

RAID also does not remove the need to plan for drive life. If you are comparing HDD and SSD endurance, see how long SSDs last. In a NAS or SAN, RAID is usually one layer inside a larger storage design; see what NAS and SAN are for that context.

Questions people also ask

Is RAID a backup?

No. RAID can keep data available after a drive failure, but it does not protect against deletion, corruption, ransomware, or controller failure. Keep separate backups.

What is the best RAID level for a home NAS?

For many home NAS systems, RAID 5 or RAID 6 is a practical balance of capacity and protection. RAID 10 is better for heavy random I/O if the budget allows.

Can RAID 0 be used for important data?

No. RAID 0 has no redundancy, so one failed drive destroys the whole array. Use it only for temporary or replaceable data.

Does RAID improve performance?

Some levels do. RAID 0 and RAID 10 improve throughput, while RAID 5 and RAID 6 add parity overhead that can slow writes.

How many drives does RAID 10 need?

At least four drives, arranged as mirrored pairs that are striped together.

Related pages

  • Data Deduplication: How It Saves Storage Space

    Data deduplication is a storage technique that removes extra copies of repeated data while keeping one unique instance. It lowers storage needs by replacing duplicates with references, and it can also reduce bytes sent across a network.

  • NVMe vs SATA: What Is NVMe and How Does It Differ?

    NVMe and SATA are storage connection protocols. NVMe is newer and optimized for SSDs and flash storage, while SATA is a legacy interface used by both SSDs and HDDs.

  • Erasure coding vs RAID: How They Compare

    Erasure coding protects data by storing parity blocks that let a system rebuild lost blocks from surviving data. RAID often stays within one host array. The better choice depends on scale and operational complexity.

  • Cloud storage classes: what they are and how they work

    Cloud storage keeps data on remote servers that you access over a network. A storage class is metadata on an object that sets its cost, performance, and access profile; providers also let you change classes manually or automate transitions.

  • SSD vs HDD: Speed, Lifespan, Use Cases

    SSD vs HDD compares flash-based solid-state storage with magnetic platter storage. SSDs are faster, quieter, and more shock-resistant, while HDDs usually offer more capacity per dollar. The best choice depends on whether speed or low-cost capacity matters more.

  • How Long Do SSDs Last? HDD and SSD Lifespan

    For most users, an SSD lasts many years and often outlasts the rest of the computer. A common range is five to ten years of normal daily use, with heavy writes shortening that span and light use extending it. Hard drives and SSDs both fail eventually, so backups remain essential.

How this page was prepared

Before writing, the topic was researched in the sources below. The page then passed automatic quality checks.

Report a mistake on this page