How to Build a Minimal ZFS NAS Without Synology, QNAP, TrueNAS (2024)

Original Article ↗ Hacker News Discussion ↗

Hardware and Form Factor Choices

  • Many use older enterprise gear (Dell PowerEdge, HP ProLiant MicroServers, HP workstations, ThinkCentre mini PCs) for ECC, iDRAC/IPMI, and many bays; downside is noise, size, and higher idle power.
  • Others prefer compact, low‑power “true home” hardware: mini‑PCs (e.g. N100‑class, Lenovo tiny PCs) or Raspberry Pi 5 with HATs; concern is limited bays and often no ECC.
  • USB multi‑bay enclosures are widely reported as flaky for important data (overheating adapters, disconnects, missing TRIM), especially with ZFS.

ZFS vs Appliances (Synology/TrueNAS/Unraid)

  • Pro‑appliance camp values: hot‑swap trays, LEDs/beeps for failed disks, integrated UI, replication, cloud sync, HA, monitoring, “don’t let me type the wrong command”.
  • DIY ZFS camp argues: ZFS itself is the “proven reliable system”, with better design than mdadm/LVM; disk replacement is usually just zpool replace.
  • Several report more data loss or compromise on off‑the‑shelf NASes than on DIY, but others distrust “hacky” setups for critical data.

ZFS on Linux vs FreeBSD/illumos

  • One side: running ZFS on Debian/Linux is “second‑class”, an out‑of‑tree module subject to kernel breakage and GPL hostility; recommends FreeBSD or illumos (OmniOS+napp-it) as most solid.
  • Counterpoint: OpenZFS targets Linux first; Linux and FreeBSD share the same ZFS code; running on LTS kernels is stable in practice. TrueNAS moving from BSD to Debian is cited.
  • Some note issues on Debian/Proxmox with NVMe under ZFS; others report long‑term success on many Linux and BSD systems.

Drives, RAID Levels, and Capacity Concerns

  • Ongoing “crisis” in disk and SSD pricing; many shuck external HDDs or buy used SAS enterprise drives for better $/TB.
  • RAIDZ1 vs RAIDZ2 vs mirrors:
    • RAIDZ2 favored to avoid data loss from read errors during resilver on large disks.
    • Some say RAIDZ1 is fine at home if you have real backups.
  • ZFS free‑space rule (keep ~20% free) is debated: some see real performance loss and fragmentation past 90%; others report near‑full pools working fine, especially with large files.
  • Consumer SSDs in RAIDZ are warned against by some due to lower endurance and small overprovisioning; others say they work but need headroom and care.

Backups, Monitoring, and Failure Handling

  • Strong consensus: snapshots and scrubs are essential; schedule regular scrubs and monitor zpool status.
  • Alerts are critical: email via smartmontools/smartctl_exporter, cron, or similar; relying only on MOTD or manual checks is seen as unsafe.
  • Guides that omit “a disk failed, here is how to detect and replace it” and alerting are criticized as incomplete.
  • Multiple tools for ZFS snapshot/backup automation mentioned (e.g. sanoid, zfs-backup), plus zfs send/receive, rsync.net, zfs.rent.
  • One user recounts losing all NAS data because they hadn’t tested recovery and lost their encryption config; “test disaster recovery” is heavily emphasized.

Alternatives to ZFS

  • Some prefer:
    • Btrfs on MDADM.
    • dm‑integrity + mdadm + XFS for a simpler, non‑ZFS stack.
    • SnapRAID + mergerfs for large mostly‑static media (more flexible disk sizes, easy drive removal).
    • Unraid for home servers; generally positive feedback for simplicity.
    • LVM with snapshots (noted to differ fundamentally from ZFS snapshots and require size management).
  • RHEL/Rocky + Cockpit, Proxmox (with ZFS and Proxmox Backup Server), and NixOS are popular bases for DIY NASes, each trading simplicity, tooling, and ecosystem differences.

AI and Storage Prices (Tangent)

  • Several posters blame current high storage and RAM prices on the AI boom; some say they would gladly give up modern LLMs to return to cheaper hardware and less “slop” content. Others see AI as transformative enough to justify the temporary pain.