← BACK TO COMPUTER MUSEUMDOCUMENTED · OCTOBER 2026

10U RACK · SMALL SERVERS · USEFUL WEIRDNESS

museum@author-lab:~$ The Homelab Rack

Private infrastructure behind the desktop

Operational 10U homelab rack
CURRENT RACKThe rack sits on a self-built cat chill/play box named “Discothèque de Chat”, which also happens to put the hardware at a much nicer working height.

Why a rack?

The HP ProCurve 1810G-24 was a gift from an acquaintance. That immediately created the obvious technical requirement: now there had to be a rack. The current cabinet is 10U.

Airflow

Two intake fans at the rear/bottom and two exhaust fans at the top, behind coarse grills that are cleaned by hand. Top exhaust is useful when cats exist in the same universe.

Power

An APC Smart-UPS SMT2200RMI2UC sits at the bottom. It protects the rack and Starlink hardware, not Cthulhu and not the UniFi PoE injectors.

Old hardware kept because it still has a job

SERVER AEDGE

OPNsense · Atom D525 · 8 GB

Supermicro CSE-502 / X7SPE-HF. WAN in, one LAN/uplink out. It provides routing, VLANs, DHCP, firewall and Unbound DNS with encrypted forwarding to Quad9.

No Suricata IDS/IPS by choice: strict segmentation and a small DNS blocklist are useful here; noisy inspection and TLS interception are not.

SERVER BSWISS ARMY KNIFE

Debian 13 · Atom D525 · 8 GB

The original Worldnode still lives. Today it handles Starlink measurements, World Observer helper feeds, the private OSCAR/ICQ service, WRP, Jellyfin, apcupsd, LocalSend receiving and an on-demand UniFi Controller.

SSH uses keys only. IPMI exists, but the cable normally does not.

SERVER CBACKUP HELPER

Debian 13 · Atom D525 · 4 GB

A small backup/helper node with a direct 40G link to Cthulhu. It exists because the old hardware is quiet, reliable and already paid for.

SERVER DPLAYGROUND

Proxmox · Xeon E3-1280 v5 · 64 GB

Supermicro X11SSL-F in a shallow 1U case. Two 1 TB WD Blue SSDs mirror the system; three more 1 TB drives form a disposable lab pool.

TempleOS, Windows 98, Solaris, Hyprland update tests — nothing here is production. If it explodes, that was probably the point.

UPS policy: safety first first first

The goal is not maximum runtime. In an outage, unnecessary systems are shut down quickly. If the HP backup server is actively copying important data, QNAP stays up long enough to remain its source while the rest of the rack sheds load.

This has been tested with the breaker deliberately switched off during a backup. QNAP and the HP MicroServer finished cleanly, and the UPS still had nearly 20 minutes of runtime left. When mains power returned, the rack recovered automatically; QNAP took its usual few minutes and announced its return with the traditional little beep melody.

Things deliberately not running

No central Grafana wallboard, no always-on UniFi Controller, no Tailscale and no remote VPN into the house. They were either tested and removed or never justified by the use case. The lab is allowed to be boring when boring is the better design.