RAID Levels Decoded
RAID turns several drives into one logical volume that is faster, safer, or both. The number tells you the trade-off between capacity, performance and how many drives can fail before you lose data. The definitions below follow Dell's PowerEdge RAID Controller (PERC) documentation.
The levels at a glance
| Level | Method | Min drives | Usable capacity | Drives it can lose | Best for |
|---|---|---|---|---|---|
| RAID 0 | Striping, no redundancy | 2 | 100% | 0 | Throughput-only scratch data |
| RAID 1 | Mirroring | 2 | 50% | 1 | Boot pairs, small critical volumes |
| RAID 5 | Striping + single distributed parity | 3 | (n−1)/n | 1 | General capacity with read performance |
| RAID 6 | Striping + dual distributed parity | 4 | (n−2)/n | 2 | Large-capacity arrays |
| RAID 10 | Striped mirrors (1+0) | 4 | 50% | 1 per mirror | Databases, VMs, write-heavy |
| RAID 50 | Striped RAID 5 groups | 6 | Varies | 1 per group | Large RAID 5 at scale |
| RAID 60 | Striped RAID 6 groups | 8 | Varies | 2 per group | Very large arrays |
RAID 0 — striping, no safety net
Dell: RAID 0 "uses disk striping to provide high data throughput, especially for large files in an environment that requires no data redundancy." All capacity is usable and it is the fastest layout, but the loss of any single drive destroys the entire set. Use it only for scratch, transcode or cache data you can afford to lose.
RAID 1 — mirroring
Dell: RAID 1 "uses disk mirroring so that data written to one physical disk is simultaneously written to another physical disk." Two drives, 50% usable capacity, survives one failure. The standard choice for OS/hypervisor boot pairs.
RAID 5 — single parity
Dell: RAID 5 "uses disk striping and parity data across all physical disks (distributed parity) to provide high data throughput and data redundancy, especially for small random access." Minimum three drives, survives one failure, gives (n−1)/n usable capacity.
Caveat for large drives: a RAID 5 rebuild must read every surviving disk in full, for hours or days on high-capacity nearline drives. With non-recoverable read errors at roughly 1 in 10¹⁵ bits, a second fault or unreadable sector during that window loses the array — which is why RAID 6 is increasingly preferred for large-capacity sets.
RAID 6 — dual parity
Dell: RAID 6 "uses block-level striping with two parity blocks distributed across all member disks," giving "protection against double disk failures, and failures while a single disk is rebuilding." Minimum four drives, survives two simultaneous failures — the safer choice for big drives where rebuilds run long.
RAID 10 — striped mirrors
Dell: RAID 10 "is a combination of RAID 0 and RAID 1, [and] uses disk striping across mirrored disks" for "high data throughput and complete data redundancy." Minimum four drives, 50% usable capacity. It delivers excellent random-write performance and fast rebuilds (the controller copies one mirror; there is no parity to recompute). The workhorse for busy databases and virtualisation hosts.
RAID 50 and RAID 60 — striped parity groups
RAID 50 stripes a RAID 0 across several RAID 5 groups (minimum six drives); RAID 60 stripes across RAID 6 groups (minimum eight). Splitting a large array into parity groups gives better rebuild behaviour and performance than one enormous parity set. RAID 50 tolerates one failure per group; RAID 60 tolerates two per group.
Hot spares
A hot spare is an idle drive the controller pulls in automatically the moment a member fails, starting the rebuild without waiting for a technician. A dedicated spare is assigned to one array; a global spare is available to any array on the controller. A hot spare shortens the dangerous "degraded" window — valuable where nobody can reach a remote site quickly. It is not a substitute for backups.
Hardware vs software RAID
| Hardware RAID | Software RAID | |
|---|---|---|
| Parity computed | On the controller's RAID-on-Chip | On the host CPU |
| Cache + power protection | On-card DRAM + battery/flash (BBU, FBWC, CacheVault) | Host RAM (needs UPS or PLP drives) |
| Boot support | Yes, transparent to the OS | Limited/varies |
| Portability | Array tied to the controller family | Array portable across hardware |
| Examples | Dell PERC, HPE Smart Array/MR, Broadcom MegaRAID | Linux mdadm, Windows Storage Spaces, ZFS |
| Cost | Controller + cache module | Included with the OS |
Hardware RAID with a healthy battery- or flash-backed write cache wins for classic block workloads and simple boot. Software and ZFS win for flexibility, checksumming and scale-out — but rely on volatile host RAM, so they need PLP-equipped SSDs or a UPS to be safe.
Choosing a level
| Workload | Recommended level |
|---|---|
| Boot / OS pair | RAID 1 |
| Busy database or VM host | RAID 10 |
| General file/app storage | RAID 5 (small arrays) or RAID 6 |
| Large-capacity nearline array | RAID 6 (or RAID 60 at scale) |
| Throwaway scratch/throughput | RAID 0 |
RAID is not backup
RAID protects against drive failure — not accidental deletion, ransomware, controller failure or site loss. Always keep a separate, tested backup.
RAID-level definitions follow Dell PERC controller documentation; usable-capacity formulas are standard.
Enterprise storage and drives, priced in rand:
Sources
- Dell PERC 11 User's Guide — summary of RAID levels
- Dell PowerEdge RAID Controller 12 (PERC 12) User's Guide
Compiled by the Server Hub newsroom from the reporting above. Every factual claim is checked against those sources before publication, and every source is linked so you can verify it yourself. How we work.
The Server Hub briefing
South African IT hardware news, once a week.
What’s new, what it costs in rand, and what it means for the kit you run — servers and storage, networking, backup power, surveillance and print. Every claim checked against a named source.