One workaround carries some sites through an outage: a phone runs a hotspot, a portable travel router turns it into one Ethernet link, and that link goes into the second WAN port of the existing gateway, holding the connection while the broadband is down. A more practical remark comes alongside it: fine for temporary use, but for something meant to stay connected, a cellular device with a data SIM is the better buy. Neither view is wrong. The difference is whether somebody keeps watching it. With a phone in the path, when the link takes over, when it steps back, and how much data is left all depend on a person. With a 5G cellular router or a 4G cellular router kept on standby year round, the device itself has to hold those decisions. The work falls into five steps, in this order.
Three placements are common. The first puts it behind the existing gateway as a second uplink only: addresses inside the site stay unchanged, nothing in the existing network is touched, and the cost is that this device carries the outbound stretch alone. The second lets it act as the site's own egress, with site devices below it, so both links switch on the same box. The third reserves one backup path for the critical devices alone and accepts that everything else goes dark. One question decides among them: does an outage need to protect the whole site or a single machine.
Port count is the hard limit behind that choice, and the one often underestimated. Two uplinks take two ports, and site devices still need somewhere to connect. The USR-G809s leaves room here: two ports configurable as WAN or LAN plus six LAN ports and two SFP fiber ports, one of which is shared with a copper port; dual Nano-SIM with the second as an embedded eSIM; RS485 and RS232 terminal blocks; DIN rail mounting; 118 x 96 x 49 mm. Running short and adding a switch means one more device, one more power feed, and one more item to sign off.
The criterion is whether the link still carries traffic, not how many bars the signal shows. A weak signal belongs in its own alarm class and does not trigger a switch, otherwise a busy cell tower produces flapping. The probe target should be the stretch this side depends on: the upstream gateway address, or the address of the platform at the far end. Picking a public address unrelated to the application means a hiccup on that host causes a needless failover.
Two checks run together. Port state answers whether the physical layer still has carrier; ICMP and heartbeat packets answer whether the far end is reachable. The USR-G816 datasheet states this plainly: automated failover and failback backup among Ethernet, cellular with dual SIM, and Wi-Fi; ICMP keepalive check and heartbeat packet check to secure availability; two SIM cards backing each other up; four Gigabit ports, one WAN configurable as LAN and three LAN; a metal case rated from -35 to 75 degrees Celsius; 125 x 103 x 45 mm; DIN rail or wall mounting. One detail is often missed: private network and VPDN SIM cards need an APN filled in. Without it, the symptom is endless registration failure, which gets misread as no coverage on site.
First, collected data. Whether there is somewhere to hold it during the outage, whether it gets backfilled afterwards, and whether that backfill happens on the device side or the platform side are questions for commissioning, not for the day of the failure. Second, long lived sessions. Once the link moves, sessions rebuild; who initiates the reconnect and how soon decides how long the gap runs. Third, the VPN tunnel. A working link is not a working tunnel, and the tunnel has to be rebuilt; tunnel reconnection is regularly slower than the link switch itself, so remote access stays down through that wait.
The USR-G806w covers the tail end of this: cellular, wired and Wi-Fi internet access switch freely, and service recovers promptly when a network exception occurs; tunnel detection keeps tunnels stable while multi-layer link detection works with automatic redial and recovery; hardware and software watchdogs cover the last layer; device offline, weak signal and traffic overrun alarms go out by email and SMS, with repeated push supported. It measures 104 x 102 x 28 mm, mounts on a DIN rail or wall, takes power from a terminal block or DC socket, and runs from -20 to 70 degrees Celsius. The number worth timing at acceptance is time to service recovery, not time to link switch.
The first few minutes after the primary link returns are often unstable, and failing back at once invites flapping. One round of flapping costs more than one switchover. The practice is to require the primary link to hold steady for a continuous period before failback, to keep the failback window clear of peak hours, and to confirm once on the spot afterwards. Switching over and switching back use the same criteria, only in opposite directions. Failback is itself a switch and has to be equally visible, as a log entry or a notification.
Long disuse is exactly what makes the day it is needed go wrong. A SIM left idle can be suspended or its plan can expire, an antenna can be moved, firmware can fall behind the rest of the batch, or the carrier can change something on its side. None of it shows up in daily use. Two cards are better off on two different carriers, otherwise one base station fault takes both paths down together.
Two things can be done. At commissioning, pull the primary link once and time the recovery, noting whether site data is retransmitting or suspended during the wait. After that, run traffic over the backup link alone once a month and read the daily and monthly traffic trend rather than the instantaneous rate. Before ordering a plan, estimate monthly usage from reporting interval, message size and point count, and set the alarm threshold at around seventy to eighty percent of the allowance. The daily and monthly traffic bar charts and live sessions on the USR-G806w exist for exactly this, and the remote management platform allows changing parameters, rebooting and upgrading firmware from elsewhere, so a confirmation does not turn into a site visit. Antenna position, firmware version and drill records belong in the handover documents; without them there is no baseline for a retest next year.
An outage gives no warning, and the value of a backup link materialises only in the minutes when it carries the load. Where it sits, whether the criteria are written down, where data goes during the switch, when to come back after recovery, and whether the idle link has been exercised: with those five steps done, a 5G cellular router or a 4G cellular router is genuinely standing by.
1. What is the fundamental difference between a 4G/5G cellular router and a regular Wi-Fi router?
Look for a SIM slot. A regular router needs broadband that already exists. A cellular router attaches to a carrier network by itself through a SIM card and turns that network into a usable egress on its LAN ports and over Wi-Fi. One more distinction is worth keeping straight: 5G is the fifth generation of cellular network, while 5 GHz is a Wi-Fi band. They are not the same thing.
2. Should the choice be 4G or 5G?
Decide by actual uplink demand, not by rated peak. Telemetry and status values of tens to a few hundred bytes every few minutes are well within 4G. Video backhaul, bulk program or firmware distribution, and several devices uploading at once are what push a decision toward 5G. Where 5G coverage is missing, a 5G model falls back to 4G on its own, so choosing 5G does not give up the 4G tier.
3. What causes rework when ordering a data SIM?
Three things. Confirm the plan is permitted inside a router, since some plans tie the SIM to a registered device or require an IMEI. Fill in the APN correctly for IoT and VPDN private network cards; when it is left empty, the symptom is that cellular never registers, which is usually misread as no coverage. And estimate monthly usage from the application, then set a threshold, rather than waiting for the bill.
4. There is no public IP on the cellular link. How can site devices be reached remotely?
Most carriers hand out private or carrier grade NAT addresses, so connections started from the public side never arrive and port forwarding is a dead end. The way out is to reverse the direction: let the site router open a VPN tunnel outward, or open its built-in web interface through the remote management platform. Neither requires an inbound port on the customer side.
5. Is going out over cellular secure?
Security is decided by the tunnel and the rules, not by the medium. The cellular stretch is the carrier's network, so data should travel inside a VPN tunnel once it leaves the site. After the tunnel is up, restrict what can be reached by address and port, rule by rule, and deny everything else by default. Compared with shared Wi-Fi, cellular removes the layer where knowing the password is enough to get in.
6. What are two SIM slots for?
Two uses. One is backup: when one card's link is unusable, the device switches to the other without changing hardware. The other is carrier diversity: coverage between carriers at the same location varies widely, and two cards on different carriers avoid a single point dead zone. At selection time, check the card form factor and whether eSIM is supported.
7. Will data be lost during the outage?
That depends on who holds the data, and it is worth asking before buying. Some designs buffer on the device side and backfill after recovery; others rely on the platform to poll again. The way to find out is to pull the primary link once during commissioning and watch whether the site retransmits, holds, or simply drops, then record the result in the acceptance notes instead of inferring it.
8. How many clients can one unit serve, and how is traffic estimated?
Client count has two separate ceilings: the wired side is bounded by port count and the address table, the wireless side by radio capacity, and the two do not add up. The wireless side is usually the one that reaches its limit first. For traffic, work out monthly volume as reporting interval tim
