RAID 0, RAID 1, RAID 5 and RAID 10 Compared
Key Takeaway
RAID 0 is about speed, not safety; RAID 1 is simple and reliable but delivers only half the capacity; RAID 5 offers high capacity efficiency but is no longer recommended on drives above 4TB, because a single unrecoverable read error (URE) during a rebuild writes off the whole array; RAID 10 has the fastest writes and the safest rebuild, at the cost of 50% of the capacity. Whichever level is chosen, RAID is not a backup.
Key Comparison
| Item | RAID 0 | RAID 1 | RAID 5 | RAID 10 |
|---|---|---|---|---|
| Minimum Drives | 2 | 2 | 3 | 4 |
| Capacity Efficiency (4 drives) | 100% | 50% | 75% | 50% |
| Fault Tolerance | 0 (any single drive failure destroys the whole array) | 1 drive | 1 drive | 1 drive per mirror pair |
| Write Penalty | ×1 | ×2 | ×4 | ×2 |
| Rebuild Speed | Not applicable | Fast | Slow (must read all drives and recalculate parity) | Fast (copies the mirror directly) |
| Risk During Rebuild | Not applicable | Low | High (no redundancy at all) | Low |
| Where It Fits | Scratch space, non-critical compute | Boot drive, small critical systems | Read-heavy, write-light cold data | Databases, virtualisation, OLTP |
Practical Takeaways
- What the write penalty actually means: how many physical I/Os a single logical write generates at the back end. A partial-stripe write on RAID 5 requires "read old data → read old parity → calculate new parity → write new data → write new parity", 4 I/Os in total, versus 2 for RAID 10 and 6 for RAID 6. Taking 8 HDDs each delivering 150 IOPS as an example, the effective write IOPS are 1,200 for RAID 0, 600 for RAID 10, 300 for RAID 5 and 200 for RAID 6.
- The key to RAID 5 failing on large-capacity drives is URE: consumer-grade SATA drives are rated at 1 in 10^14 bits (roughly once per 12.5 TB read), while enterprise-grade SATA/SAS is 1 in 10^15 bits (roughly once per 125 TB). During a rebuild the controller must read every block of every surviving drive in full, and a URE on any one of them means the array can never complete the rebuild.
- Quantifying the risk: taking a RAID 5 array of 6 x 4TB consumer-grade drives as an example, the rebuild must read about 20TB while the URE rate is one per 12.5TB - the odds of completing the rebuild successfully are not encouraging. The larger the drives, the longer the rebuild (at the 8TB class it can reach 24 to 72 hours), and the higher the risk.
- Capacity decision: with drives of 2TB or less for non-critical use, RAID 5 is still technically workable; with drives of 4TB or more, RAID 6 is recommended instead (dual parity, tolerating two drive failures); for databases and virtualisation, go straight to RAID 10. SSD/NVMe arrays carry a lower URE risk, but the RAID 5 write penalty still applies.
- What RAID cannot protect against: ransomware encryption, accidental deletion by staff, fire, flooding, theft, logical corruption and file system errors. The right approach is the 3-2-1 rule (3 copies of the data, 2 different media types, 1 off-site or offline); for ransomware protection, an offline or immutable backup copy is the only reliable insurance.
- The battery trap in hardware RAID: if Write-Back caching is enabled without BBU or supercapacitor protection, cached data is lost on a power cut and the array may be left inconsistent. At procurement you must confirm that a BBU or Flash-backed Cache (supercapacitor) is included, and monitor its health regularly - a lithium battery BBU typically needs replacing every 3 to 5 years, while a supercapacitor / Flash-backed Cache module is rated for 5 years or more and generally needs only health monitoring.
Tip: Deployment practice: use RAID 1 for the boot drive; RAID 10 for databases and virtualisation; RAID 6 or RAID 60 for large-capacity cold data and backup repositories. Configure at least 1 hot spare (Hot Spare); keep 15% to 20% of usable space free; and do not buy all drives from the same batch (drives from the same batch have a chance of failing at the same time).
