Internet Failover for Home Offices: Why a Second Path Belongs in the Design

Most home offices already have a backup: if the internet goes down, switch to the phone’s hotspot. For email and browsing during an outage, that is often enough. A phone hotspot can get several devices online. During a live call, three things work against it.

It is manual. Someone has to notice the outage, turn the hotspot on, move each device over, and rejoin the meeting, while the other party watches a frozen frame.

The office network stays behind. Devices that join the hotspot leave the office network. Wired equipment such as a desk phone, a printer, or a wired desk does not move with them, and how many devices a hotspot accepts depends on the phone and the carrier.

It depends on the cellular signal at the desk. Where reception indoors is weak, the hotspot is weak too.

An installed backup line takes the person out of that sequence. The gateway notices the failure and moves the whole office network to the second line on its own. A call or VPN session in progress may still need to reconnect; what changes is that nobody has to rebuild the connection by hand.

The design has four parts, and each is a decision rather than a product.

A managed dual-WAN gateway. On our projects this is a UniFi gateway with two WAN ports. It runs health checks against each path on a fixed interval and applies a policy when the primary stops answering: which traffic moves, how quickly, and how recovery back to the primary is handled so the system does not flap between paths.

An independent second path. Usually Starlink on a professional mount, sometimes a second terrestrial provider. The word that matters is independent: a backup that shares the primary provider’s infrastructure fails with it.

Detection thresholds matched to the work. The gateway is told what counts as failure: how much latency and loss to tolerate, for how long, before switching. A threshold tuned for a telemedicine desk is tighter than one tuned for a household that mostly streams.

A commissioning test. Configured behavior is not the same as verified behavior. Before handoff, the primary path is disconnected on purpose, with the applications the office actually uses running, and what recovered is written into the handoff record. Some sessions survive a WAN change without a visible interruption; some applications, including certain VPN and authenticated sessions, reconnect rather than continue. Commissioning is where that is learned, not the first real outage.

Five documented projects, each with its own published record, showing what each system did when it was tested at handoff. None of these records holds a switchover time, so none is given. Readings are from commissioning day on specific lines and properties, not a promise about yours.

Two internet paths do nothing if the gateway that chooses between them has lost power. In Leesburg, the problem the client felt most often was not the ISP. It was the utility flicker that rebooted the whole stack and cost five minutes every time.

So on these projects the core goes on a pure sine-wave UPS chosen for the installed load: gateway, switching, and the access points that serve the office. Runtime is finite and depends on that load and on the battery’s age. At Partlow, the metered power distributor shows the rack drawing about 193 watts across five outlets; that is the kind of figure a battery has to be matched to, and anything plugged in later shortens the runtime. None of these records includes a measured runtime.

The office workstation is hardwired wherever the property allows. A wired desk is not exposed to the wireless variables that a call on a laptop across the house is, and it is the one link in the chain that has to be certain.

A continuity system the owner cannot read is a liability the day the installer is unavailable. Every project above closed with a written record: the installed equipment, the topology, the configured gateway policy and thresholds, what the commissioning test showed, the power assumptions, and how support is reached.

The Lorton residence was commissioned without the owner on site, so the record had to stand in for a walkthrough. It became a 20-page infrastructure guide written for the family and for whoever services the system next: which outlet restarts the gateway, which device not to touch, what expected behavior looks like versus a fault. Partlow received an owner guide and a quick-reference card, with every device and port verified against the management console before the record was closed.

That is what serviceable means. The system can be understood, restarted, and extended by reading it, not by reverse-engineering it.

Each home-office backup system is quoted in writing after a free first conversation, and the price follows the property: how long the core must stay up on battery, whether the workstation is hardwired, whether a wall-mount rack organizes the equipment, and whether the continuity backbone covers one office or a main house and its outbuildings.

Provider charges remain separate from the project. The primary ISP and the backup provider continue billing under their own plans, and no recurring maintenance contract is required with the installation. The scope, the commissioning standard, and the handoff are described on the Home-Office Reliability page.

It is when one person’s work depends on the connection, the office is at home, and a dropped call costs more than the inconvenience.

When a small team depends on the connection, the same parts are scoped for the office instead (see the small-office note). And when the property has a guest house, a barn, or a detached office that also has to stay online, the office becomes one part of an estate network and the failover design becomes a property decision rather than a desk decision.

Where the second path is Starlink, the mount, the sky view, and the cable path decide whether the backup is dependable. We cover that separately in why Starlink fails without proper installation, and the choice between Starlink and a cellular second path in Starlink vs. T-Mobile Home Internet.