What an older network shutdown does to proxy hardware
A line on my second shelf spent a long stretch failing every evening. It would start misbehaving around seven and heal itself before eleven, and during the bad window everything reported healthy. The modem said it was registered. The port kept answering. Requests through it crawled or gave up halfway. I swapped the cable, then moved the modem onto a different hub, then moved the sim into a neighbouring modem, and the fault stayed with the modem, which is normally where a hunt like that ends.
It took another week before I looked at the one field I had never bothered recording: the radio technology the modem claimed to be attached to. In the afternoon it reported the current network. At seven in the evening it reported a generation older. The line had been leaning on an old network at exactly the hours anyone tried to use it, for as long as I had owned it.
That gap between what a modem claims and what a modem does matters more than it used to, because carriers retire older network generations. Retirement is an ordinary part of running a mobile network. It has happened before and it will happen again. Each time it does, every modem that has been quietly treating the old generation as a safety net loses the net, and nothing announces the loss.
A fallback nobody configured
A modem attaches wherever its configuration points it, and loyalty has nothing to do with it. It is told to prefer one network, and it is permitted to drop onto something older whenever the preferred one is weak, busy or briefly out of reach. That permission ships enabled on almost everything you can buy, and for a phone it is the right default. It is why a call survives a lift and why data keeps moving in an underground car park.
On a shelf full of modems the same default means part of your fleet is standing on a network you never chose, at hours you cannot predict, and you have no idea which part unless you go and look. The modem logs no warning. The software above it sees a registered device holding an address, and a registered device holding an address is all most monitoring has ever asked for.
Why a single test tells you nothing
You test a line once, usually at your desk, usually in the afternoon, when the tower serving your building is half asleep. The line reports the network you paid for, you write that down, and the note becomes a permanent fact about the line for the rest of its working life.
The dependency lives at the edges. The corner of the room where the signal sits one bar lower. The evening hours when a few million people get on trains and start watching video. The thirty seconds after a reconnect, when the modem is choosing where to land and takes whatever answers first. A line can spend ninety percent of its life on the network you believe it is on and still fail every time you actually need it to hold.
The four shapes a retirement takes
A retirement never arrives as an outage. Lines going dark together on a tuesday morning would be easy to see. What arrives instead is a fleet that turns unreliable in ways that read exactly like ageing hardware, and in my experience it takes four shapes.
A line that fails to come back after a drop. It was fine, something interrupted it, and it never re registered. It sits powered and visible, holding no connection, and a power cycle brings it back for a day.
A registration loop. Attach, hold for a minute, drop, attach again. The customer sees a session that breaks every few minutes and blames the port. From the rack the device looks busy and alive the whole time, because it is.
Throughput that collapses at the worst hours. The line attaches and works, and the numbers are a third of what it used to do, in the evening, when the customer is running the job that pays for the line.
Rotation that stops being dependable. You ask for a fresh address, the modem drops the connection, and instead of coming straight back it wanders for thirty or forty seconds before settling. Some days it never settles.
None of those shapes mention a network generation, and every one of them looks like a bad cable or a tired modem, which is how a slow retirement turns a whole rack into a maintenance problem nobody can name. The tell is distribution. If the trouble tracks one network rather than one shelf, one hub or one batch of cables, the rack is fine and the answer sits upstream of you.
What this means if you are buying
For somebody who has never touched a modem and never wants to, the relevant fact is that the cheapest listings in this market ride on the oldest hardware, because that is the only way the price works. Devices bought years ago on a spec sheet that was already dated, kept alive because they still pass a health check, sold on price because price is all there is to sell on.
A retirement is a slow, predictable and entirely visible way for that inventory to fail, and the seller gets no say in the timing. There is no dramatic day when such a seller goes offline. Support gets slower. More of your lines need swaps. The replacements come off the same shelf as the originals, so each swap buys you a fortnight. Quality drifts down over months while the price stays exactly where it was, because the price was the whole product.
So ask a supplier a boring question and listen to how it comes back. What are the modems. Do they hold the current network with fallback switched off. When was hardware last replaced. An operator who runs his own kit answers in a sentence and probably complains about the cost while doing it. A reseller cannot answer at all, because the devices belong to somebody he has never met.
Auditing the hardware you already own
If you run your own rack, this is a couple of afternoons of work, and it is worth doing before you need it.
Log the reported radio technology on every device, on a schedule, and keep the history. Every five minutes is plenty. Read it over a week rather than once, because the whole point is that a single reading proves nothing. What you want at the end is a count per device: how many samples on the current network, how many on the old one. A device that never once touched the old network can be crossed off the worry list. A device that spends a fifth of its evenings down there has a date on it.
Log signal quality alongside, because the pair tells you whether a fallback is about the device or about where you put it. A modem that drops on a strong signal has a radio problem. A modem that drops only when the signal sags has a placement problem, and placement is cheap to fix.
Then run the honest test. Lock a device to the current network so it has nowhere to fall back to, and leave it locked through a full day and a full evening. A device that holds is a device you can keep. A device that spends the evening unregistered has been carrying a dependency you never priced in, and now you know its name. Do the rotation test in the same locked state, because the ugliest results show up there. A modem that attaches cleanly on lock and then cannot rotate cleanly on lock is telling you its recovery path ran through the old network the whole time.
Do all of it per device, never per batch. Two modems from the same order, same firmware, thirty centimetres apart on the same shelf, will give you different answers. Band support gets cut differently between production runs, antennas vary, and a spec sheet is a marketing document.
What to do with the ones that fail
Throwing them out is the expensive answer and usually the wrong one. A marginal device is still useful once you know it is marginal. Put it where the signal is best. Give it work that tolerates a wobble. Keep it away from anything holding a long session or a logged in account.
Then plan the replacement on your own calendar. Buy a few at a time, spread across months, so you are never replacing an entire fleet in one panicked week at whatever price that week happens to offer. A fleet bought together fails together, and that lesson only needs learning once. Stagger the models too, because a single new device type across a whole rack means a single new set of surprises across a whole rack.
What I am not claiming
The reported technology field is what the modem says about itself. Modems lag, modems round things off, and some report the network they are camped on rather than the one carrying your bytes. Treat it as a good signal rather than a measurement.
Locking a device can also make things worse. In a genuinely weak spot, a modem with nowhere to fall back to gives you nothing at all, where before it gave you something slow. That trade is fine on a shelf you control and wrong in a location you do not.
And a retirement cannot be rehearsed, because the old network is still there while you test. Everything above is a way of finding out which of your devices quietly depend on it, so that the day it changes is a tuesday you already planned for.
The batch I got wrong
I once bought a run of very cheap modems on a spec sheet that listed the current network. Every unit attached to it on the bench, so I racked them and stopped thinking. They turned out to be the worst devices I have ever run, and it took most of a year to work out why. Their band support was thin. The spec sheet said the right network, and in practice they could reach it on only a couple of frequencies, so any tower load, or any wall between them and the antenna, sent them straight down a generation.
They were never faulty. They did precisely what they were built to do, at a price that said exactly what they were, and I read the headline number on the box instead of the table underneath it. The fix cost more than decent hardware would have in the first place, which is the oldest lesson in this trade and evidently one I needed to pay for twice.
If you would rather rent lines from someone who logs what his radios are actually attached to and replaces hardware before it becomes your problem, that is what I run at singaporemobileproxy.com.
Get new guides and videos first — join the Telegram channel.