Understanding RAID Configurations for Hosting Servers

What RAID Actually Is

RAID (Redundant Array of Independent Disks) combines multiple physical drives into a single logical unit. The goals: performance (faster read/write), redundancy (survive drive failures), or both.

When a hosting provider says "RAID 10 NVMe storage," they're describing how your data is distributed across physical drives. The RAID level affects performance, durability, and usable capacity.

The RAID Levels

RAID 0 — Striping (Performance Only)

How it works: Data is split (striped) across all drives. A 128KB file is split into chunks and written across all drives simultaneously.

Drive 1: [Block A] [Block C] [Block E]
Drive 2: [Block B] [Block D] [Block F]

Performance: Excellent. Reads and writes are parallel across all drives. 4 drives × 500 MB/s = ~2,000 MB/s potential throughput.

Redundancy: None. If ANY drive fails, ALL data is lost.

Usable capacity: 100% of total drive capacity.

Real-world: Almost never used in hosting. The performance gain doesn't justify the complete lack of redundancy. One drive failure = complete data loss.

RAID 1 — Mirroring (Redundancy Only)

How it works: Data is written identically to two (or more) drives. Every write goes to both drives; every read can come from either.

Drive 1: [File A] [File B] [File C]
Drive 2: [File A] [File B] [File C]

Performance: Good reads (can read from either drive). Writes are as fast as the slowest single drive (must write to both). Negligible performance gain over a single drive.

Redundancy: Excellent. Can survive failure of all but one drive in the mirror set. With 2-drive RAID 1, one drive can fail — data remains accessible.

Usable capacity: 50% of total drive capacity. 2 × 2TB drives = 2TB usable.

Real-world: Common for small setups (2 drives) where redundancy matters and capacity is sufficient. The simplicity is appealing — if one drive fails, the other has a complete copy.

RAID 5 — Striping with Distributed Parity

How it works: Data is striped across all drives, with parity (error-correction) information distributed across drives. Parity allows reconstruction of data if any single drive fails.

Drive 1: [Block A] [Block B] [Parity C]
Drive 2: [Block C] [Parity B] [Block D]
Drive 3: [Parity A] [Block E] [Block F]

Performance: Good reads (parallel across all drives). Writes are slower (must calculate and write parity). Small random writes are particularly slow (read-modify-write cycle).

Redundancy: Can survive failure of exactly 1 drive. If a second drive fails during rebuild, data is lost.

Usable capacity: Total capacity minus 1 drive. 4 × 2TB drives = 6TB usable.

Real-world: Historically common, now declining. The "RAID 5 write hole" (inconsistent parity during power loss) and long rebuild times on large drives make it risky for modern high-capacity drives. A 16TB drive takes 24+ hours to rebuild — during which a second failure is catastrophic.

RAID 6 — Striping with Dual Parity

How it works: Like RAID 5, but with two independent parity blocks. Can survive two simultaneous drive failures.

Performance: Reads similar to RAID 5. Writes are slower (two parity calculations). Small random writes are even slower than RAID 5.

Redundancy: Can survive failure of any 2 drives.

Usable capacity: Total capacity minus 2 drives. 6 × 2TB drives = 8TB usable.

Real-world: Used when RAID 5's single-parity risk is unacceptable. The dual parity provides a safety margin during long rebuilds. Still declining in favour of RAID 10 for most hosting workloads.

RAID 10 (1+0) — Mirroring + Striping

How it works: Data is first mirrored (RAID 1 pairs), then striped (RAID 0) across mirror pairs. Best of both worlds: performance of striping, redundancy of mirroring.

Mirror Pair 1:  Drive 1: [Block A] [Block C]    Drive 2: [Block A] [Block C]
Mirror Pair 2:  Drive 3: [Block B] [Block D]    Drive 4: [Block B] [Block D]

Performance: Excellent reads and writes (parallel across mirror pairs). No parity calculation overhead.

Redundancy: Excellent. Can survive at least 1 drive failure (all drives in the other mirrors). Can survive up to N/2 failures if they're in different mirror pairs.

Usable capacity: 50% of total drive capacity. 4 × 2TB drives = 4TB usable.

Real-world: The standard for hosting servers. Fast, resilient, simple to understand and recover. The 50% capacity overhead is the tradeoff.

Non-Standard RAID

RAID 50, 60

Combinations of RAID 5 or 6 with RAID 0. Multiple RAID 5 arrays striped together = RAID 50. Better performance than pure RAID 5/6, same redundancy characteristics.

RAID-Z (ZFS)

ZFS's software RAID implementation. RAID-Z1 ≈ RAID 5, RAID-Z2 ≈ RAID 6, RAID-Z3 ≈ RAID 7 (triple parity). ZFS avoids the RAID 5 write hole with copy-on-write semantics. Increasingly used instead of hardware RAID on modern hosting servers.

Hardware RAID vs Software RAID

Hardware RAID:

• Dedicated RAID controller card handles all RAID operations
• OS sees a single logical drive
• Offloads RAID calculations from CPU
• Battery-backed cache improves write performance
• Controller failure can make data recovery difficult (must find identical controller)

Software RAID (mdadm, ZFS, Btrfs):

• OS handles RAID operations using CPU
• No special hardware needed
• Portable — drives can be read on any Linux system
• Modern CPUs handle RAID calculations with negligible overhead
• More flexible (ZFS snapshots, compression, deduplication)

For most modern hosting setups, software RAID (especially ZFS) is superior to hardware RAID. The CPU overhead is negligible, the portability advantage is significant, and ZFS's data integrity features (checksumming, self-healing) provide protection against silent data corruption that hardware RAID cannot match.

What Hosting Providers Actually Use

Budget VPS (1-2 drives):

• RAID 1 (two SSDs/NVMes mirrored)
• Simple, reliable, 50% capacity loss

Mid-range VPS/Dedicated:

• RAID 10 (four NVMe drives — 2 mirror pairs striped)
• Best balance of performance and redundancy
• Standard for quality hosting

Enterprise SAN:

• Proprietary RAID-like systems
• Multiple layers of redundancy
• Hot spares, automated rebuilds, predictive failure detection

Cloud providers (AWS, GCP, Azure):

• Don't use traditional RAID (they use distributed storage systems like Ceph or proprietary equivalents)
• Data is replicated across multiple machines, not just multiple drives

What RAID Level Should Your Host Use?

If a hosting provider doesn't disclose their storage configuration, ask. "NVMe storage" without specifying RAID level could mean a single unprotected drive.

RAID and Backups — Not the Same Thing

This is the most important thing to understand about RAID: RAID is not a backup.

RAID protects against drive failure. It does not protect against:

• Accidental file deletion (rm -rf will remove files from all RAID drives simultaneously)
• Filesystem corruption
• Malware/ransomware
• Server theft or physical destruction
• Application errors that corrupt data
• User error (overwriting the wrong file)

RAID + off-site backups = real data protection. RAID without backups = you'll lose data eventually, it just won't be from a single drive failure.

The 2021 OVH Strasbourg fire was a vivid demonstration: RAID 10 arrays across dozens of servers were useless when the physical building burned. Customers with off-site backups recovered. Customers who relied on RAID alone lost everything.

Quick Reference

For hosting servers in 2026: RAID 10 (hardware or ZFS mirrors+stripes) is the standard. RAID 5 is outdated. RAID-Z2 with ZFS is excellent if you're building your own server. And always, always have backups.