In this Field Note
A remote power cycle can take seconds. Getting a person to the equipment may involve a drive, a gate code, a site contact, and an afternoon that had other plans.
That is the best argument for replacing a row of PoE injectors with managed power. A tidier cabinet is pleasant. Being able to recover one stuck radio without dispatching someone is an operating capability.
It works only if the power system is compatible, the port map is correct, and you can still reach the control that removes the power. The last condition is where some very neat installations become surprisingly unhelpful.
Trace the recovery path before choosing a switch
Picture a site with one backhaul radio, several access radios, and a managed PoE switch. This is a design example. If an access radio hangs while the backhaul and switch remain reachable, a remote per-port cycle may solve the immediate problem.
If the backhaul hangs and it is your only route into the site, the same button is now on the other side of the outage. The switch has not lost its capability; you have lost access to it.
Write down the fault you expect to recover and follow the management route under that fault. An independent management path may be justified at an important site. At a smaller one, accepting a visit for a backhaul failure may be reasonable. Either is better than buying a recovery feature under the assumption that it can reach itself.
Keep administrative access restricted and credentials controlled. A power-control interface can interrupt service just as effectively as the router configuration screen.
Check that “off” actually removes power
Disabling an Ethernet port does not necessarily switch off its PoE output. Verify a real, documented per-port power operation on the exact model and software you plan to use.
For example, MikroTik’s PoE-Out documentation distinguishes interface power settings and a timed power-cycle operation. It also describes model-dependent voltage and power limits. The existence of those controls on one product says little about another product with a similar front panel.
Test the operation on compatible spare equipment before production use. Confirm how long power remains off, whether the setting persists, what is logged, and whether management returns. A timed cycle is operationally different from switching a port off and hoping you remain connected long enough to turn it on again.
Label ports against live device identities. Record the radio, management address, cable label, service dependency, and approved power mode. An old label can be more dangerous than a missing one because it encourages confidence.
Build the electrical schedule one device at a time
“PoE” is not a complete compatibility specification. Passive and negotiated power arrangements differ. Match the powered device and source using their actual requirements: voltage, polarity, powered pairs, negotiation, current, and maximum demand.
Do not force power onto an unknown device to see whether it starts. A compatible connector is not an electrical approval.
| Record for every port | Why it belongs in the schedule |
|---|---|
| Exact device and hardware revision | Similar product names can conceal different power requirements. |
| Supported input and source mode | Establishes compatibility before power is applied. |
| Maximum demand and startup allowance | Idle consumption does not describe every operating condition. |
| Cable length, type, and condition | The installed path affects delivered power and Ethernet behavior. |
| Port, bank, and total source limits | An acceptable individual load can still overload a shared budget. |
| Priority and recovery role | Documents what must remain available to restore the rest of the site. |
As a planning example, four devices specified at 18 W total 72 W at those stated measurement points. That does not establish that a supply labeled 72 W is suitable. The design must account for how the manufacturer specifies delivered power, cable losses, source consumption, startup behavior, and the applicable operating margin. Avoid adding percentages to incompatible numbers and calling the result a power budget.
Verify the complete cold start under approved conditions. Moving devices onto a live switch one by one does not exercise the same startup demand as restoring power to the whole site.
Grounding, bonding, surge protection, enclosures, and electrical installation still require qualified treatment. A managed switch does not make outdoor cabling or mains work a software problem.
Keeping the injectors may be the right answer
A network-controlled PDU with individually switched outlets can provide remote recovery while retaining compatible injectors. That can be useful where devices have particular power requirements or the existing arrangement is otherwise sound.
Compare the complete alternatives: switch and supplies versus PDU, injectors, adapters, outlet count, rack space, monitoring, and spares. Verify actual switching capability; a metered PDU may report consumption without controlling individual outlets.
Also compare what fails together. Consolidating radios behind one switch introduces a shared data and power dependency. Some models preserve PoE during a software reboot, but that does not protect against loss of the supply or chassis. Test the actual behaviors you are relying on.
A stable site with one or two injectors may not justify conversion. At a larger site, the value often comes from better remote diagnosis as much as avoided trips: link state, error counters, and supported per-port power readings can help distinguish a dead device from a bad Ethernet path.
Be suspicious of a radio that needs regular rescuing
Before cycling power, confirm identity and capture useful evidence. Repeated reboots may conceal a cable, thermal, power, or firmware fault. Convenient recovery should make the investigation easier, not replace it.
I generally leave ping watchdog disabled. I use it when a device has a known failure mode that a power cycle clears and the automated recovery is justified. A missing ping alone does not identify a hung radio. The probe target, routing, or upstream path may have failed while the radio is perfectly healthy.
Where automation is warranted, allow for startup, require sustained failure, control retries or provide a cooldown, and log and alert on each action. Test loss of the probe target and upstream path. A site-wide reboot party is still an outage, even if every invitation was sent automatically.
If the available watchdog cannot express the required safeguards, use manual recovery or a controller that can. The firmware rollout policy matters here too: a recurring software fault deserves a controlled fix, not an indefinite schedule of power cycles.
Leave a bypass with the people who need it
Keep a saved configuration, compatible spares, and a documented way to restore the critical path. A known-good injector for the backhaul can remain useful after the rest retire.
Move one service at a time, verify traffic, and update the schedule as the work proceeds. Demonstrate remote power control, loss of management, whole-site recovery, and the agreed bypass before closing the job.
Oaktown Tech can build the compatible bill of materials, port schedule, and recovery procedure for a site conversion. You should leave knowing which faults can be handled remotely, which still require a visit, and what the person making that visit needs to bring.

Join the discussion