Reference Build
The Rack Lab
The Rack Lab is for people who want to learn on the same kind of equipment they meet at work: rack servers with management controllers, a dedicated storage server, 10 GbE and a properly isolated management network. It idles around 467 W and belongs in a basement.
- 467 WMeasured idle at the wall
- 4,091 kWhEnergy per year
- 93.1 TiBComfortable storage at 80%
- AudibleBasement or utility room only
Who it is for
The Rack Lab is an enterprise training ground that also runs the household. It is the right build if you want hands-on experience with management controllers, 10 GbE storage networking, large ZFS pools and multi-host virtualization, and if you have a basement or utility room where noise and heat are acceptable. Read The Real Cost of a Free Server before you start; this build is designed to be as efficient as rack equipment reasonably gets, and it still costs real money to run.
Parts by role
| Rack | 18U four-post open-frame rack, adjustable depth set for 800 mm equipment, on a level floor away from water heaters and drains |
| Firewall | 1U short-depth x86 appliance with 10 GbE SFP+ and 2.5 GbE ports, running an open-source firewall distribution |
| Switching | 24-port managed PoE+ core switch, an 8-port 10 GbE SFP+ aggregation switch with direct-attach copper cables, and a separate small switch used only for management controllers |
| Wireless | Three ceiling access points |
| Compute | Two 2U short-depth servers with server boards, management controllers, 16 cores, 128 GB ECC memory, mirrored NVMe SSDs and dual 10 GbE |
| Primary storage | 2U twelve-bay storage server with an HBA in pass-through mode, 64 GB ECC memory, twelve 16 TB CMR drives as two six-wide RAIDZ2 vdevs, and a mirrored SSD pool for VM storage over NFS |
| Backup | 1U four-bay server with four 20 TB drives in RAIDZ2, pulling replication nightly with its own credentials |
| Utility and console | Single-board computer for monitoring and scheduled power control, a KVM-over-IP device with an 8-port KVM switch, and USB serial adapters for the firewall and switches |
| Cameras | Six wired PoE cameras on their own VLAN, recorded by a container on one compute server to a dataset on the storage server |
| Power | 2U line-interactive UPS around 2200 VA with a network management card |
| Approximate cost | $11,000 to $16,000 at late-2026 prices, highly dependent on drive prices and how much is bought used |
Measured power
| Device | Idle at the wall |
|---|---|
| Firewall | 22 W |
| Core PoE+ switch, before PoE loads | 30 W |
| 10 GbE aggregation switch | 14 W |
| Management switch | 4 W |
| Three access points | 18 W |
| Two compute servers, tuned firmware profiles | 156 W |
| Storage server, twelve drives spinning | 135 W |
| Backup server | 38 W |
| Utility computer and KVM-over-IP | 8 W |
| Six PoE cameras | 27 W |
| UPS self-consumption | 15 W |
| Total | 467 W, about 4,091 kWh a year |
At 17 cents per kWh that is about $695 a year, and the rack produces roughly 1,590 BTU per hour of heat. Powering the second compute server down for eight hours each night saves about 228 kWh a year; the utility computer sends power-on commands to its management controller every morning.
Network plan
All eight segments of our example plan are in use, with two refinements. Management controllers connect to the dedicated management switch, which reaches the rest of the network only through the firewall's management interface, and only from the admin workstation and the utility computer. Storage traffic between the storage server and both compute servers runs over the 10 GbE aggregation switch on the non-routed storage VLAN. See Out-of-Band Management and VLANs and Segmentation.
What it runs
- Everything the Closet Lab runs, spread across two hosts with plenty of headroom.
- A dedicated learning area: a small container orchestration cluster built from VMs, a log aggregation server, a network security monitoring sensor on a mirrored switch port, and disposable VMs for testing automation.
- A staging environment that mirrors the household services, used to test updates before they reach production.
Storage and backups
Two six-wide RAIDZ2 vdevs of 16 TB drives provide 128 TB of usable space, about 116.4 TiB, and survive two failures in each vdev. Keeping below 80 percent leaves about 93.1 TiB of comfortable space. Check alternative layouts with the Storage Capacity Calculator.
- Copy 1: the primary pool with tiered snapshot schedules per dataset.
- Copy 2: the backup server pulls encrypted replication nightly. It holds 40 TB of usable space, enough for every irreplaceable and costly-to-recreate dataset with months of versions.
- Copy 3: raw encrypted replication of the irreplaceable tier to a reciprocal partner's NAS over a mesh overlay, plus cloud object storage for photos and documents.
Running it well
- Every server's firmware power profile is set to balanced, unused add-in cards are removed and fans are on automatic.
- All configuration lives in a git repository and is applied by configuration management, so either compute server can be rebuilt from scratch in under an hour.
- A yearly restore drill rebuilds one service from the off-site copy alone. See The Restore Nobody Tested.