A small-office stack that survives Belize: pfSense, UniFi and Proxmox

The constraints come first

Every office network I have built in Belize starts from the same three facts. Power is not continuous. Internet is not continuous. And if a piece of hardware dies, the replacement is not arriving tomorrow, it is arriving whenever it clears customs.

That shapes the design more than any feature list does. I want a stack I can reason about at two in the morning over a phone hotspot, made of parts I can re-provision from a config file rather than a memory of what I clicked. For a small office of ten to thirty people, the combination I keep coming back to is pfSense at the edge, UniFi for access and switching, and Proxmox for anything that has to run locally.

pfSense at the edge

pfSense does the work that actually matters when the link degrades. Two WAN interfaces, one on the primary fibre or cable service, one on a secondary provider or an LTE modem, with a failover gateway group. The important part is not the failover itself but the monitoring: pick a sensible monitor IP, tune the loss and latency thresholds, and test the cutover deliberately rather than discovering it during a busy morning.

The rest of what I set up on it is unglamorous and pays for itself:

• VLANs separated at the firewall, not just at the switch, with rules written to be readable a year later
• A site-to-site or road-warrior VPN so I can reach the network without opening anything to the public internet
• DNS resolver on the box, with local overrides so internal names work whether the WAN is up or not
• Scheduled config backups pulled off site, because a firewall you cannot rebuild in twenty minutes is a firewall that will cost you a day

I keep the ruleset small. Clever firewall configurations are the ones nobody can maintain after the person who wrote them moves on.

UniFi for access and switching

UniFi earns its place because handover works, the switching is manageable, and the office manager can look at a client list without me. Access points go where the coverage survey says, not where the ceiling tile is convenient. Guest traffic sits on its own VLAN with no route to the internal networks. PoE from managed switches means one less wall wart to lose during a surge.

I run the controller as a virtual machine on the Proxmox host rather than on a cloud key. Fewer physical things to fail, and snapshots before firmware upgrades.

Proxmox for everything else

Proxmox holds the workloads that should not depend on a link staying up: file services, the UniFi controller, a local database or reporting instance, a small application server, sometimes a Windows VM for accounting software that refuses to live anywhere else. On the QBridge work, having a stable local host to run the bridge from was the difference between a reliable push and a nightly guessing game.

Backups go to local storage and then off site. Snapshots before every change. A UPS that Proxmox actually talks to, so the host shuts down cleanly instead of pulling the plug on a running database. Grounding and surge protection are not optional here, and I would rather spend on that than on a faster CPU.

What it buys you

None of this is exotic. It is three well documented pieces of software, configured so that the recovery path is written down and the failure modes are boring. That is the whole goal.

If you are planning an office move or replacing something that has been limping along, the contact form on this site reaches me directly.