What RAID actually solves
RAID (Redundant Array of Independent Disks) combines multiple physical disks into a single logical volume to improve performance, fault tolerance, or both. It is not a backup strategy — RAID protects against hardware failure of individual disks but does not protect against accidental deletion, ransomware, file corruption, or site-level disasters. Always pair RAID with a 3-2-1 backup plan.
Quick selection cheat sheet
- Maximum speed, zero fault tolerance: RAID 0 (striping only, any disk failure destroys the array).
- Simple redundancy for 2 disks: RAID 1 (mirror, 50% capacity efficiency).
- Balanced capacity and redundancy: RAID 5 or RAIDZ1 (single parity, lose one disk worth of capacity).
- Stronger redundancy for large arrays: RAID 6 or RAIDZ2 (double parity, survives two simultaneous disk failures).
- High IOPS plus redundancy: RAID 10 (mirror of stripes, half raw capacity).
- Very large parity pools: RAID 50 or RAID 60 (stripe of RAID 5/6 groups).
RAID 0 — Striping
RAID 0 splits data evenly across all member disks. Sequential read and write throughput scales nearly linearly with disk count, making it attractive for scratch space, video editing caches, or temporary build artifacts. However, if any disk fails, the entire array is lost. RAID 0 should never hold data you cannot afford to recreate.
RAID 1 — Mirroring
RAID 1 writes identical data to two (or more) disks. Read performance can improve since requests can be served from either disk, but write performance is limited to the slowest member. Capacity efficiency is 50% — half of your raw storage is consumed by the mirror. It is a common choice for boot drives and small but critical datasets.
RAID 5 / RAIDZ1 — Single parity
Both RAID 5 and ZFS RAIDZ1 distribute parity across all disks. You lose one disk worth of usable capacity in exchange for the ability to survive a single disk failure. They were the workhorse for medium arrays for many years, but with modern multi-terabyte drives, rebuild times can stretch into days, during which a second failure would be catastrophic. For arrays with disks larger than 4 TB, consider RAID 6 or RAIDZ2 instead.
RAID 6 / RAIDZ2 — Double parity
RAID 6 and RAIDZ2 use two parity blocks per stripe, so the array can survive two simultaneous disk failures. This is the recommended baseline for arrays of 6 or more disks, or any pool using disks 4 TB or larger. The trade-off is slightly slower writes (two parity calculations) and two disks worth of capacity lost to parity.
RAID 10 — Mirrored stripes
RAID 10 (also written as RAID 1+0) creates a stripe of mirrored pairs. It offers excellent random read and write IOPS, predictable rebuild times (one disk copied from its mirror), and tolerates multiple disk failures as long as no mirror pair fails entirely. The cost is 50% capacity efficiency. It is the preferred layout for transactional databases and high-IOPS workloads.
RAID 50 and RAID 60 — Striped parity groups
RAID 50 and RAID 60 stripe data across multiple RAID 5 or RAID 6 sub-groups. They are designed for very large arrays where a single parity group would have impractical rebuild times. The exact capacity and fault tolerance depend on how the groups are configured, so always verify the topology with your storage vendor or model it before purchase.
RAIDZ1, RAIDZ2, RAIDZ3 — ZFS parity
ZFS uses its own parity levels: RAIDZ1 (single parity, like RAID 5), RAIDZ2 (double parity, like RAID 6), and RAIDZ3 (triple parity, no direct hardware RAID equivalent). RAIDZ benefits from ZFS features like end-to-end checksums, copy-on-write, and snapshots, which provide stronger data integrity guarantees than traditional hardware RAID.
Rebuild risk and modern drive sizes
As individual drive capacity has grown to 18 TB and beyond, the time required to rebuild a failed drive in a parity-based array has grown with it. During a rebuild, every remaining drive is read end-to-end, increasing the chance of an unrecoverable read error (URE). This is why double-parity layouts (RAID 6, RAIDZ2) are now the recommended baseline for arrays larger than 4 disks, regardless of drive size.