All RAID Types Explained: A Practical Guide for NAS and Homelab

Updated: 2026-05-12

TL;DR

There are roughly nine RAID families that matter today. Most home NAS owners end up with one of four: RAID 1 for a two-bay box, RAID 6 or RAIDZ2 for four to eight drives, SHR-2 on Synology, and Unraid's parity model for mixed-size pools. RAID does not protect you against fire, theft, ransomware, or accidental deletion — that is what a backup is for. This page covers what each level does, what it costs in capacity, and which mistakes show up most often.

All RAID levels at a glance

“All but one drive's worth” means: 5×10 TB in RAID 5 gives 40 TB usable (one drive's capacity becomes parity). Detailed formulas and minimum drive counts: methodology.

Before you pick a level

These five questions decide the answer more than the RAID level itself.

How many drives, and what sizes? A 2-bay always ends up RAID 1 or single-parity SHR/Unraid. A 4–6 bay is where the RAID 5 vs 6 vs RAIDZ2 debate actually matters. Mixed sizes rule out classic RAID and push you toward SHR or Unraid.

Do you have a backup? RAID protects against drive failure. It does not protect against ransomware, fire, theft, a power surge that fries every disk, or accidentally deleting the wrong folder. Without an off-site backup, no RAID level is safe.

How large are the drives? Modern 16–24 TB drives take 24–48 hours to rebuild from parity. The probability of a second drive failing during that window grows with array size. The widely-cited rule that RAID 5 stops being safe around 8–12 TB drives comes from this math — see the rebuild calculator for the numbers on your setup.

What is the workload? Plex and Jellyfin streaming barely stress any RAID. VMs, databases, and live video editing want RAID 10 or a ZFS mirror layout. Cold archives are happy on SHR or Unraid where idle drives spin down.

Which OS? Synology gear gets SHR. Unraid gets Unraid parity. TrueNAS gets ZFS (RAIDZ or mirrors). Stock Linux or Windows gets mdadm or hardware RAID. The OS often picks the level for you.

RAID 0 — striping only

Data is split into chunks and written across all drives in parallel. Twice the drives roughly means twice the throughput. There is zero redundancy: lose one drive and the array is gone, because each disk holds only half of every file. RAID 0 belongs on scratch disks, editing caches, and game installations that can be re-downloaded. Nowhere else.

RAID 1 — mirror

Every drive holds an identical copy. With two drives you can lose one. With three, you can lose two. Usable capacity equals the smallest drive no matter how many you add — the rest is redundancy. The classic two-bay NAS configuration: simple, safe, and the rebuild after a failure is just a file-by-file copy from the surviving drive. Backblaze's Drive Stats reports failure rates on the order of 1% per year for typical NAS drives,¹ so on a two-drive mirror the chance both fail in the same week is small — but not zero. A backup is still required.

RAID 5 and RAIDZ1 — single parity

One drive's worth of parity is spread across all members. You can lose any one drive and rebuild. Usable capacity is (N−1) × smallest drive.

Single parity is mathematically fine on three or four small drives. It gets risky as drives grow because rebuilds take longer, and during a rebuild the array has no further fault tolerance. A second drive failure — or even one unrecoverable read error (URE) on a surviving drive — ends the array.

A consumer WD Red Plus 12 TB is rated for one URE per 10¹⁴ bits read.² Reading every byte of a 5-drive RAID 5 made of 12 TB drives during rebuild scans about 44 TB, which is on the same order of magnitude as the URE threshold. Whether you actually hit one is luck — but the trend line argues against single parity on large drives. More on RAID 5 →

RAIDZ1 has the same fault tolerance as RAID 5 but adds per-block checksums, so silent corruption is detected and (with redundancy available) repaired. It does not change the underlying single-parity risk profile.

RAID 6 and RAIDZ2 — double parity

Two drives' worth of parity. Survives any two simultaneous failures. Usable capacity is (N−2) × smallest drive.

This is the default recommendation for arrays of five or more drives, or anywhere drives are 8 TB and up. The extra parity drive is the cost of being safe during a multi-day rebuild. RAIDZ2 applies the same idea inside ZFS, where the checksumming also catches bit rot during a scrub. More on RAID 6 →

RAID 10 — stripe of mirrors

Drives are paired into mirrors, then the mirrors are striped. Always an even drive count. Half the raw capacity is redundancy. Reads scale with the number of mirrors; writes scale less because every write touches both halves of a pair.

The win is rebuild speed: replacing a failed drive copies one mirror block-by-block from its surviving partner. There is no parity calculation and no whole-array scan. Hours instead of days. That makes RAID 10 the right pick for VMs, databases, and any workload where downtime risk during a rebuild matters more than capacity efficiency.

RAIDZ3 — triple parity

ZFS only. Survives any three simultaneous failures. Usable capacity is (N−3) × smallest drive. RAIDZ3 makes sense for very large pools (10+ drives) where rebuild windows are long and a third failure during recovery is no longer hypothetical. For typical 4–8-drive home arrays, RAIDZ2 is enough.

SHR-1 and SHR-2 — Synology Hybrid RAID

SHR is Synology's adaptive parity layout. Internally it creates multiple RAID groups across the size tiers of your drives, then combines them into one volume. Effect: mixed drive sizes contribute their full capacity, not just the size of the smallest drive.

Example: 3×8 TB plus 1×4 TB in SHR-1 gives roughly 20 TB usable. The same drives in classic RAID 5 give 12 TB. SHR-1 has the fault tolerance of single parity; SHR-2, of double parity.³

SHR is only available inside Synology DSM. It cannot be moved to TrueNAS, Unraid, or stock Linux without reformatting. Synology's hardware-compatibility policy changed in 2024–2025: certain newer Plus-series models restrict full feature parity (including SHR) to Synology-branded drives.⁴ If you plan to use third-party drives, check the compatibility list before buying. More on Synology DSM →

Unraid — independent disks plus parity

Unraid takes a different approach. Each data drive is an independent ext4, XFS, btrfs, or ZFS filesystem with its own contents. One or two additional parity drives hold the XOR of all data drives, sized at least as large as the biggest data drive.

A single data-drive failure is rebuilt from the parity. A failure of a data drive together with the parity drive only loses the contents of that one data drive — the others remain readable on their own. Striped RAID would lose everything in the same scenario. The trade is raw speed for resilience and flexibility.

Mixed drive sizes are natural. Adding a drive does not require rebalancing — assign it, pre-clear it, done. Unraid is paid software (one-time license tiers from 49 to 249 US dollars as of 2024).⁵ More on Unraid →

JBOD

Drives concatenated end-to-end as one volume. No redundancy, no striping. Loss of one drive loses only the files that lived on that drive — unlike RAID 0, where loss is total. JBOD is for scratch space and append-only archives backed up elsewhere. Most NAS operating systems offer it as an option; few people use it long-term.

Pick by use case

2-bay home NAS, family photos: RAID 1 (or SHR-1 on Synology). Simple, two-drive failure window, easy to recover.

4-bay home NAS, all same size: RAID 6 / RAIDZ2 / SHR-2 if drives are 8 TB or larger; RAID 5 / RAIDZ1 / SHR-1 is acceptable for smaller drives in non-critical scenarios.

4–8-bay mixed drive sizes: SHR-2 on Synology, Unraid with two parity drives elsewhere.

Plex / Jellyfin media server: Capacity matters more than IOPS. SHR-2, RAIDZ2, or Unraid 2P are all reasonable choices.

VMs, databases, video editing: RAID 10 or a ZFS pool of mirrors. Rebuild speed and IOPS beat capacity efficiency.

8+ drive pool, ZFS: Two RAIDZ2 vdevs (e.g. 6+6) usually beat one 12-wide RAIDZ3 on rebuild time and IOPS.

Common mistakes

Treating RAID as a backup. RAID protects against drive failure. It does not protect against malware, accidental deletion, whole-NAS hardware failure, fire, or theft. A 3-2-1 strategy (three copies, two media, one off-site) is the minimum.

Single parity on 12 TB+ drives. Rebuild windows of one to three days plus URE math make RAID 5, RAIDZ1, and SHR-1 progressively riskier as drives grow. Move to double parity.

Mixing SMR and CMR drives. Shingled-Magnetic-Recording drives can drop out of a rebuild because write performance collapses under sustained load. Most NAS-rated drives are CMR; verify the spec sheet.⁶

Ignoring URE risk during rebuild. Consumer drives are rated for one unrecoverable read error per 10¹⁴ bits. Pro and enterprise drives are typically 10¹⁵, an order of magnitude better. The rebuild calculator shows the cumulative risk for your array.

Assuming btrfs RAID 5/6 is safe. The btrfs RAID 5/6 implementation has had known data-loss bugs for years and is still considered experimental as of kernel 6.x.⁷ Use ZFS RAIDZ, mdadm RAID, or Synology SHR if parity is required.

Frequently asked questions

Is RAID a backup?

No. RAID provides uptime when a drive fails. A backup is a separate copy of your data that survives the failure of the entire system. You need both.

What is the difference between RAID 1 and RAID 10?

RAID 1 mirrors all drives into a single copy, so usable capacity equals one drive no matter how many you add. RAID 10 pairs drives into mirrors and then stripes the pairs, so usable capacity is half the total raw capacity.

Can I add drives to an existing RAID array?

It depends on the system. Synology SHR and Unraid support drive-by-drive expansion. Classic RAID needs a reshape or rebuild. OpenZFS 2.3 added RAIDZ expansion in 2024, so ZFS can also grow a vdev by one drive at a time.

Hardware RAID or software RAID?

For NAS use cases in 2026, software RAID (mdadm, ZFS, btrfs, Synology, Unraid) is the default. Hardware RAID cards add a single point of failure and tie the array to a specific controller model.

Does RAID slow my computer down?

On a modern CPU, parity calculation is free. Writes touch every drive in the array, so absolute write IOPS can be lower than a single drive, but for streaming, archival, and most home workloads this is not noticeable.

Can I use SSDs in RAID?

Yes. The same RAID levels apply. SSDs have different failure modes than HDDs, and most consumer SSDs are not endurance-rated for sustained array writes. Datacenter or enterprise SSDs are a better fit.

What is silent data corruption?

Bits that flip on disk without the controller flagging an error. Classic RAID cannot detect it. ZFS and btrfs add per-block checksums; if a checksum mismatch is found during a scrub and redundancy is available, the bad copy is repaired automatically.

How long does a RAID rebuild take?

Roughly six to twelve hours per TB on consumer NAS hardware. ZFS pools only resilver used blocks but throttle to limit production impact. Use the rebuild calculator to estimate based on your drive size and array width.

Sources

1. Backblaze Drive Stats — quarterly drive failure rates across 250,000+ drives, published under CC-BY 4.0 at backblaze.com/cloud-storage/resources/hard-drive-test-data.
2. WD Red Plus product brief, Western Digital, documents.westerndigital.com.
3. Synology Knowledge Center, “What is Synology Hybrid RAID (SHR)?”, kb.synology.com.
4. Synology hardware-compatibility policy update for 2025 Plus-series models, Synology product pages.
5. Unraid pricing, unraid.net/pricing.
6. CMR vs SMR drive disclosure, Western Digital, blog.westerndigital.com.
7. Linux kernel btrfs documentation, RAID 5/6 status, btrfs.readthedocs.io.

Further reading

RAID 0 Explained: Speed Without a Safety Net

RAID 5 vs RAID 6: Which Should You Actually Pick?

RAID for Home Users: Everything You Actually Need to Know

RAID 10 vs RAID 6: Which One Fits Your NAS?