What is a RAID array? RAID explained simply

RAID in one minute: the simple version

Imagine you have four identical drives. You can use them four ways. Add them up and write each file in pieces across all four at once: you get all the space and lots of speed, but if any one drive dies, every file is missing a piece (RAID 0). Pair them up so every drive has a twin holding the same data: you lose half the space but any single drive can fail (RAID 1 / 10). Or keep a checksum — a little extra maths spread across the drives that can recalculate whatever was on a dead disk: you lose only one drive’s worth of space and survive one failure (RAID 5), or lose two drives’ worth and survive two failures (RAID 6). That is the whole idea; everything else is detail.

RAID (Redundant Array of Independent Disks) joins multiple physical drives into a single logical volume. Depending on the ‘level’ you choose, the array can be faster, larger, more fault-tolerant, or a blend of the three. None of it is magic — every benefit is paid for with capacity, redundancy or both. Understanding that trade is the key to choosing well.

The common RAID levels

RAID 0 (striping) splits data across all drives for maximum speed and full capacity, but with zero redundancy — lose any one drive and the entire array is gone. It is for scratch space and throughput, never for data you care about. RAID 1 (mirroring) writes identical copies to two drives; you keep half the raw capacity but survive a full drive failure, and reads can be faster. RAID 5 stripes data plus distributed parity across three or more drives, surviving one drive failure while sacrificing only one drive’s worth of capacity to parity — an efficient balance. RAID 6 adds a second parity block, surviving two simultaneous failures at the cost of two drives’ capacity — the safer choice for large arrays. RAID 10 (a stripe of mirrors) combines RAID 1’s redundancy with RAID 0’s speed, keeping half the capacity but offering fast rebuilds and strong performance.

Usable capacity: what you actually keep

Redundancy is not free — it is paid in capacity. A RAID 1 mirror of two 8 TB drives gives 8 TB usable, not 16. A five-drive RAID 5 of 8 TB disks gives about 32 TB usable (four drives’ worth, one consumed by parity). RAID 6 of the same five drives gives about 24 TB (two drives consumed). Because most parity levels limit usable space to the smallest member, mixing capacities wastes the surplus on larger drives. To plan a build precisely — usable space, parity overhead and the effect of drive count — use the capacity calculator before you buy.

How much space do you lose with RAID?

Here is the same set of four 8 TB drives (32 TB raw) under each level. The space you ‘lose’ is not wasted — it is what buys the redundancy.

Two things shrink the number further. First, drive makers count a terabyte as 1,000 GB while most operating systems count 1,024, so an 8 TB drive shows as about 7.3 TB. Second, the filesystem and NAS software reserve a few percent. The overhead also changes with drive count: RAID 5 on three drives loses a third of raw space, but on eight drives only an eighth — which is why parity RAID gets more efficient as arrays grow.

Hardware RAID, software RAID and NAS flavours

You no longer need a dedicated RAID card. Software RAID — Linux mdadm, Windows Storage Spaces, macOS Disk Utility, and the RAID built into every home NAS — runs on the main CPU and is the right default for home and small-office use. ZFS (TrueNAS, Proxmox) uses its own names: RAID-Z1 is roughly RAID 5, RAID-Z2 roughly RAID 6, and mirrors are mirrors — with the bonus of checksums that detect silent corruption. Synology’s SHR and Unraid’s parity array let you mix drive sizes without wasting as much space. Hardware RAID cards still exist in servers, but tie your array to one controller model: if the card dies, you need a compatible replacement to read your data.

Can you mix RAID levels on one system?

Yes. RAID levels apply per array, not per computer, so one machine or NAS can run a RAID 1 mirror of two SSDs for the operating system or apps and a separate RAID 5 or 6 array of hard drives for bulk storage. What you cannot do is put the same drive in two arrays (some software allows partitions to be split between arrays, but it complicates recovery and is best avoided). Within one array, keep drives identical in capacity and recording technology — mixing sizes wastes the surplus and mixing CMR with SMR drives invites rebuild trouble.

Fault tolerance and the rebuild-risk problem

Surviving a drive failure is only half the story; the dangerous moment is the rebuild. When you replace a failed drive, the array reconstructs its data by reading every remaining drive in full — an intense, hours-to-days workload precisely when the array has no remaining redundancy (in single-parity RAID 5). If a second drive fails or hits an unrecoverable read error during that rebuild, the whole array is lost. With today’s very large drives, rebuild times are long and this risk is real, which is why many builders now choose double-parity RAID 6 for big arrays so a second failure during a rebuild is survivable.

Two factors make rebuilds worse: slow drives and shingled drives. An SMR drive can stall a rebuild for days or fail it outright, and a drive without time-limited error recovery can be dropped mid-rebuild — both reasons to use proper NAS or enterprise CMR drives in any array.

So what should you actually run?

For a two-drive home setup, RAID 1 is simple and safe. For a small NAS of three to five drives where you want efficient capacity, RAID 5 is reasonable — with backups. For larger arrays, or any array built from large drives where rebuild times stretch into days, RAID 6 (or equivalents like ZFS RAID-Z2) is the prudent default. For performance-critical workloads, RAID 10 trades capacity for speed and fast rebuilds. Whatever you choose, pair it with an independent backup following the 3-2-1 strategy — and stock the array with enterprise or NAS-grade CMR drives.