ADR-004 — pi-box: add storage, or replace the machine?

    Status: Accepted
    Date: 2026-06-14
    Resolution (2026-06-14): Keep the Pi; ordered a Fikwot FX815 256GB 2.5" SATA SSD (£34.99) + UGREEN 2.5" UASP USB-3 caddy (£9.99) = £44.98. Reusing the internal WD Blue was rejected — it's an internal sata drive holding live data, so freeing it means opening the desktop for no net saving over a £45 new drive.
    Context: pi-box (Raspberry Pi 4, 8GB RAM, bought ~£100) is meant to be the cheap, stable, low-maintenance box that just runs low-resource apps (e.g. the SparkyFitness calorie tracker) — explicitly not a dev box. That role belongs to streaming-server (the Beelink N100). pi-box felt sluggish; diagnosis (dev-log 2026-06-14) found the only weakness is its SD card (~750 random-write IOPS, plus the long-term wear/corruption risk SD cards carry). The question this ADR answers is not "which SSD" — it's "is upgrading this Pi worth it for a run-only box, or should I spend similar money on a different machine?"

    Decided & ordered

    Keep the Pi; add an SSD. Ordered: Fikwot FX815 256GB 2.5" SATA SSD + UGREEN 2.5" UASP caddy = £44.98. Not a new machine — the Pi is already adequate for the run-only role; its sole gap is storage, and the main reason to fix it is reliability (SD-card wear), not speed. SSD power draw is fine on the Pi 4's USB (unlike the spinning HDD, which isn't).

    The real question: speed isn't the issue — reliability is

    Why the SD card matters

    less than the diagnosis implied — but still matters The painful symptom (slow nixos-rebuild, sluggish SSH) is a dev workload — thousands of small writes. This box won't do that. For running SparkyFitness (≈500MB RAM, occasional small database writes) the SD card's ~750 write IOPS is tolerable. So the upgrade isn't really about speed for this use case — it's about reliability: SD cards are the #1 cause of Raspberry Pi failure, and a database writing continuously will wear and eventually corrupt one. An SSD removes that failure mode.

    Options

    OptionCostVerdict for a run-only box
    A. Keep SD card£0❌ Works, but the wear/corruption risk is exactly what you don't want in an unattended always-on box. False economy.
    B. Reuse WD Blue SSD + M.2-SATA enclosure~£16–27❌ Rejected. The WD Blue is internal (sata) and holds live data — freeing it means opening the desktop, for no saving over a £45 new drive.
    C. New 2.5" SATA SSD + UASP caddy~£45✅ Chosen — Fikwot FX815 256GB (£34.99) + UGREEN UASP caddy (£9.99). Clean, in stock, leaves the WD Blue untouched.
    D. Used thin client (Dell Wyse 5070 / Lenovo M-series)~£40–90🟠 Strictly better hardware (x86, onboard SATA/NVMe, 8GB+), similar money — but means a fresh NixOS install + migration + finding a good unit. Only worth it if you want to retire the Pi.
    E. New N100 mini PC~£130–160❌ Overkill for run-only, and the most money. The Pi already has the RAM/CPU headroom this role needs.

    Decision

    Proposed: Option B if the WD Blue can be freed (£20), otherwise Option C (£57). Keep the Pi.

    The £100 already spent on the Pi is sunk. The live decision is marginal: ~£20–57 to make the box you own reliable, vs ~£40–160 to replace a box that is already adequate (plus migration effort). For a stable, low-resource, run-only role, the Pi 4 8GB is genuinely enough — CPU, RAM, power and thermals all measured healthy; storage is the only gap, and a cheap SSD closes it.

    When to revisit (i.e. when replacement

    would be justified)

    • The apps outgrow ~8GB RAM or want real multi-core x86 grunt.
    • You want to consolidate onto one machine and retire the Pi.
    • The Pi's USB power budget proves flaky under the SSD (then a powered USB hub or a thin client with internal storage wins).

    Consequences

    • ✓ Keeps cost minimal (~£20–57) and reuses owned hardware.
    • ✓ Removes the SD-card reliability/wear failure mode — the real win for an unattended box.
    • ✓ No OS reinstall or migration; the existing NixOS pi-box config carries over.
    • ✗ Still a Pi 4 — if future apps get heavy, this only delays a replacement, doesn't avoid it.
    • ✗ Option B depends on the WD Blue being spare and buying the correct enclosure type (see the M.2/SATA/NVMe explainer in the dev-log — easy to buy the wrong one).