Tested Devices

Tested Devices — the community catalogue

✅ Live catalogue. This is the wiki face of docs/HARDWARE.md. It is short, and that's honest — one unpaid person can't buy every board. If you run it somewhere, adding a row is the single most useful contribution you can make.

Legend: ✅ works · ⚠️ works with caveats · ❌ no · ❓ untested (expected, unverified)

The catalogue

Device	Arch	RAM	Profile	Status	CAKE ceiling	Notes	Reporter
Raspberry Pi 2B	armv7	1 GB	legacy-sbc	⚠️	n/a (not inline)	Sinkhole / print / firewall / time / monitor only. No onboard RTC → add a DS3231.	maintainer
dev/CI: WSL2 Debian	x86-64	—	—	✅ authoring	—	Configs authored here; shellcheck / sh -n clean. Idle Virgin baseline ~12 ms / <1 ms jitter / 0 % loss (2026-07-15).	maintainer
Raspberry Pi 4 / 5	aarch64	2–8 GB	modern	❓	~600–900 Mbit	Expected to work. Not yet tested.	—
Intel N100 (2× i226-V)	x86-64	8–16 GB	modern	❓	~1–2.5 Gbit	The intended inline shaper. Not yet tested — hardware not yet purchased.	—

Read those ❓ rows literally. The N100 is the box this whole design points at, and it has never been run. The shaper configs are reviewed and shellcheck-clean, but they are not hardware-proven. If you get there before the maintainer does, your report is genuinely valuable.

What "tested" means here

Deliberately strict, because an optimistic catalogue is worse than a short one:

• ✅ works — you ran it, it did its job, you'd leave it running.
• ⚠️ works with caveats — it runs, but something needs saying (a limit, a workaround, a gotcha). This is the most useful status.
• ❌ no — you tried; here's the wall you hit. Also useful — it saves the next person.
• ❓ untested — nobody has run it. An educated guess, clearly labelled.

A ⚠️ with a real caveat is worth more than an optimistic ✅. If it only half works, say so — that's the honest thing, and it's what stops someone wasting a weekend.

Report your device

Open a Discussion or send a PR against docs/HARDWARE.md using this template:

- Device / SoC:
- Arch / RAM / storage:
- Profile (legacy-sbc / modern):
- OS + version:
- What worked / what didn't:
- CAKE bufferbloat grade (if inline):  before → after
- Gotchas / notes:

If you ran the shaper inline, the before → after bufferbloat grade is the number this project most wants. Grab it from waveform.com/bufferbloat on a wired machine.

Known-good and known-bad components

Thing	Verdict	Why
Intel i226-V 2.5GbE	✅ recommended	The reference NIC for the inline path.
Realtek 2.5GbE	⚠️ avoid	Driver pain — the classic wasted evening.
DS3231 RTC (I²C, ~£3)	✅ recommended on any RTC-less Pi	Without it the 2B forgets the time on every power-off, which quietly breaks TLS and DNSSEC.
Jetson (any)	❌ wrong tool	The GPU can't run CAKE and the A57 CPU is weaker than a Pi 4's.
Pi 2B as a gateway	❌ no	100 Mbit NIC on the shared USB 2.0 bus. Superb sidecar though.

The profiles, briefly

• legacy-sbc — Pi 2B / armv7 / ≤1 GB / Alpine. Sinkhole, print, firewall, DDNS, chrony, monitor. No Wolfi (no armv7 target), no ClamAV (RAM), no local metrics store, not inline.
• modern — Pi 4/5 aarch64 or x86 N100. Everything above plus Wolfi bases, CAKE shaping (N100, inline), SELinux/AppArmor, and a metrics store.

Full detail: docs/PROFILES.md.