← sources

My Tiered Homelab Storage Topology: ZFS Mirrors + MergerFS/SnapRAID on Proxmox VE

source: reddit (reddit:homelab)published 2026-09-21 07:23ingested 2026-09-22 02:39

retained text

Hey everyone,

Sharing the storage topology diagram for my primary Proxmox node ("Tesseract"). I spent a few months planning this layout to balance data integrity, power efficiency, and hardware constraints, and it's been running rock-solid and self-sustaining for a couple of months now.

The Problem I Wanted to Solve:

I didn’t want a single, monolithic ZFS RAIDZ2 pool for everything. Having 6 spinning disks wake up and burn power just to stream a single episode on Jellyfin felt wasteful, and I wanted to be able to mix drive sizes and filesystems without risking an entire array if multiple drives failed.

The 4 Storage Tiers:

Hot Tier / AppData (/rpool): Mirrored 512GB NVMe drives handling Proxmox OS, LXC root filesystems, and databases. High random IOPS, low latency.

Critical Data Tier (/tank): Mirrored 4TB Seagate IronWolf drives. Used for personal files, family photos (Immich), documents (Paperless-ngx), and Time Machine backups. Real-time ZFS checksumming and self-healing.

Bulk Media Tier (/mnt/eternity): A unified FUSE pool via MergerFS backed by SnapRAID for scheduled parity.

Combines mismatched drives: two 4TB IronWolfs (single-disk ZFS), one 2TB Barracuda (ext4), and a fast 2TB Samsung 970 EVO Plus (ext4).

Protected by a dedicated 4TB Barracuda parity disk (/mnt/hangman).

Why this setup? Only the specific drive hosting the active media file spins up; the rest stay asleep. If catastrophic multi-drive failure strikes, only the data on the failed drives is lost; the surviving drives remain 100% readable standard filesystems.

Ephemeral Churn Tier (/scratch): A dedicated 500GB SSD for torrent downloads, extraction, and staging. Keeps write fragmentation and churn off the spinning disks and saves NVMe write endurance.

Hardware Specs:

Host: Repurposed gaming rig — AMD Ryzen 5 5600X, 32GB DDR4 RAM (ZFS ARC hard-limited to 8GB in modprobe.d).

HBA: LSI / Broadcom 9300-8i SAS HBA (IT Mode) handling all the spinning rust and SATA drives directly.

GPU: AMD Radeon RX 6600 XT (shared to Jellyfin and Immich containers via /dev/dri/renderD128 cgroup mapping).

Backups: Off-host Proxmox Backup Server (PBS) running on a dedicated Raspberry Pi 5 with an NVMe drive.

Mount dependencies in /etc/fstab use [x-systemd.requires=zfs-import.target](http://x-systemd.requires=zfs-import.target) to ensure MergerFS only initializes after the underlying ZFS pools are imported.

Happy to answer questions about the mount order, SnapRAID automation scripts, or container permissions!

document fetched 2026-09-22 00:37 · canonical https://reddit.com/r/homelab/comments/1wm5tx0/my_tiered_homelab_storage_topology_zfs_mirrors