OpenZFS
OpenZFS · cloud storage · reviewed by Stan K. · Source: data snapshot · data as of 2026-11-12
The layer underneath it all — checksums, snapshots, RAID-Z. It never speaks S3; it is what your NAS or object gateway stands on.
#6 of 12 in Cloud Storage
ToolScore breakdown — why 64/100
10 criteria × 10 points. Low pillars are visible on purpose — that's the point of a score you can audit. Methodology →
What it does well
- ▸Checksums, snapshots, RAID-Z — the integrity foundation everything above stands on
- ▸zfs-2.4.4 shipped 21.08.2026 — steady maintenance on the base layer
- ▸CDDL: a per-file weak copyleft with no network-service obligations
- ▸Composable: it is what your TrueNAS or your object gateway stands on, whatever S3 server you pick
Where it costs you
- ▸It never speaks S3 — no API, no protocol, no service semantics
- ▸ARC cache defaults to ~50% of system RAM — set an explicit limit on small hosts
- ▸Avoid dedup — the hub's one-word warning for a feature that eats RAM for breakfast
- ▸Scored strictly as an object-storage pick it lands low (64): it is the layer beneath, not the store
Hidden traps — the stuff the pricing page doesn't say
Out of the box, OpenZFS's ARC cache will claim roughly half of system RAM. On a small host sharing duties with containers or an S3 gateway, that default starves everything else — set an explicit limit on small hosts. And avoid dedup: the RAM cost is the trap.
OpenZFS appears in the hub's table because everything else stands on it — TrueNAS runs on it, object gateways store on it. But it never speaks S3: no endpoint, no protocol. Buying a disk and creating a pool is the start of a storage stack, not the end.
Know the limits before you pay: read the traps above, then decide with eyes open.
Visit official siteAffiliate link where available — never affects the score.
- official OpenZFS — CDDL license and zfs-2.4.4 (checked 2026-11-12 via the hub's GitHub pass) verified
Data as of 2026-11-12. Prices and limits change monthly — report anything stale. How we verify: methodology.
Closest alternatives
Enterprise scale (PB): RBD + CephFS + RGW (S3 + Swift) in one system — real operational weight, too
Flat and predictable: 1TB storage + 1TB egress in one hourly-billed base capped at €6.49 — S3-compat
Home/office NAS on OpenZFS — object access arrives via apps (e.g. the MinIO app), not built in; SCAL
OpenZFS review
In-depth analysis · last updated 2026-11-12
The floor of the stack
OpenZFS is in the hub's object-storage table for taxonomy, not competition: it is the layer underneath it all — checksums, snapshots, RAID-Z — and it never speaks S3. Your TrueNAS runs on it; your object gateway stores on it. The hub scores it strictly as an object-storage pick, which is why it sits mid-table rather than at the top: it is the foundation, not the store.
What it gives every row above
The integrity primitives that make self-hosted storage defensible are here: checksums that catch silent corruption, snapshots that make deletes survivable, RAID-Z that makes disk loss routine. zfs-2.4.4 shipped 21.08.2026 — steady maintenance on the base layer. CDDL is a per-file weak copyleft: no AGPL-style obligations at the service boundary, which is why everyone from appliances to gateways builds on it freely.
The two tuning sentences
The hub's field advice fits in two lines: ARC cache defaults to ~50% of system RAM — set an explicit limit on small hosts. And avoid dedup. Both traps are RAM-shaped; both bite hardest exactly where self-hosters run — small boxes doing several jobs at once.
How the pillars land
Track record 8: zfs-2.4.4 on 21.08.2026 — the base layer of a large ecosystem, maintained on schedule. Feature set 7.5: checksums, snapshots, RAID-Z — the deepest integrity toolset in the table. Real limits 7 and Privacy 7: your disks, your filesystem, no meter and no third party. Pricing transparency 7: $0 under CDDL with the traps documented in the open. Price/Value 6.5 and Ease 6: free but expert-tuned — ARC limits and dedup avoidance are not optional knowledge. Transfer speed 6, neutral: no object-transfer facts apply. S3/API interop 3: it never speaks S3 — the category's core requirement, absent by design. Support & docs sits at the neutral 5 — nothing SLA-like verifiable in this snapshot.
Verdict — NICHE, 64/100
You do not choose OpenZFS instead of an object store — you choose it underneath one. Set the ARC limit, skip dedup, and let SeaweedFS or Garage speak S3 on top of the pool.