SKYHOUSE.dev Journal

Why: Storage is currently scattered across internal SATA, two USB "easystore" externals, and a hot 2-bay "toaster" — eating all the fast USB ports and with no unified redundancy. Consolidating into one 8-bay DAS with a parity pool simplifies management, frees USB ports, and gives real drive-failure protection for the "friends & family" services (and a neighbor's offsite backup). This entry is the plan (hardware ordered; not yet executed).

Hardware chosen

• Mediasonic ProBox H82-SU3S3, 8-bay USB 3.0 (5 Gbps) DAS, "Like New" from the Mediasonic Store (~$229). Chosen over the QNAP TL-D800C ($499) on value — 5 Gbps is already 2× the 2.5 GbE LAN and 5× the ~0.93 Gbps WAN, so USB speed is a non-issue here. (The Promise Pegasus2 R8 was rejected: Thunderbolt-2, won't connect to this PC.)
• Known quirks to expect (free returns, so burn-in first): stock fan can be noisy (80×80×20mm; Noctua/Evercool swap if it carries into the adjacent TV room); no per-drive trays/LEDs (tray-less slot-load — label each drive's slot with its serial); occasional USB drive drop-outs reported (SnapRAID tolerates this; use a good cable + burn-in before trusting it).

Redundancy: SnapRAID + mergerfs (parity pool)

• mergerfs pools all data drives into one namespace; SnapRAID adds parity drive(s) for failure protection. Fits mostly-static media, mixed drive sizes, and piecemeal growth — and replaces the old mirror-pair + rsync `_backup` scheme (more space-efficient).
• Hard rule: the parity drive must be ≥ the largest data drive. With 14 TB data drives, parity must be 14 TB+. So the 6 TB/4 TB drives are always data, never parity.
• Best placement: internal SATA is the most reliable connection (no port-multiplier, no USB-dropout risk) — put parity drive(s) on internal SATA, bulk read-mostly media in the DAS.
• Neighbor's offsite backup: dedicate 1–2 bays, encrypted at rest (LUKS) so his data is private on our hardware.

Drive shuffle

• Shuck the two WD easystore 14 TB externals into the DAS. Rationale: USB ports are the scarce resource, not bays — shucking moves 2 port-hogs onto the DAS's single shared port, freeing 2 fast ports. ⚠️ 3.3V-pin gotcha: these Ultrastar-class drives may not spin up in a backplane that feeds 3.3 V on SATA power pin 3 (PWDIS) — if a shucked drive stays dark, cover pin 3 with Kapton tape (or use a Molex→SATA adapter).
• Move the 2 "toaster" drives into the DAS (they overheat in the toaster; the DAS has active cooling).
• Retire sdb (plex2_backup) — SMART-degraded; do not reuse it in the DAS.
• Add the 6 TB + 4 TB from the retiring Windows PCs only after their useful data is backed up. Strategy: fill the DAS with drives on hand, then replace the smallest/oldest in pairs over time (ideally as drive prices ease).
• The "toaster" stays as an ad-hoc swap slot for temporary drives.

USB port layout (4 fast 10 Gbps Type-A ports)

The board's USB is Intel "Gen 2x1" = 10 Gbps; four rear Type-A ports, no USB-C. Hubs time-share bandwidth (active device gets full speed; only simultaneous transfers contend), so:

• Port 1 → DAS (dedicated; it wants full bandwidth during SnapRAID syncs).
• Port 2 → a USB 3.x powered hub (NOT USB 2.0 — that would cap a flash drive at 480 Mbps) routed to the front: flash drives stay fast when active, plus printer / keyboard / mouse.
• Port 3 → toaster (only when needed). Port 4 → spare.
• A dedicated fast port only matters for two storage devices transferring hard at once — which this workload (one stream + occasional copy) doesn't do.

Execution checklist (when the DAS arrives)

1. Load drives, label each slot with its serial.
2. smartctl --scan + per-drive smartctl -a — confirm SMART passes through the enclosure.
3. Burn-in: sustained read/write for a day or two, watch dmesg for resets/drop-outs, before trusting it with the only copy of anything (esp. the neighbor's backup).
4. Install mergerfs + SnapRAID; assign parity (≥14 TB, ideally internal SATA) and data drives; schedule nightly snapraid sync + periodic scrub.
5. Set up the neighbor's LUKS-encrypted bay(s) + his push target.