How long a mobile IP should actually last

Someone asked me last month whether our sticky window was fifteen minutes or thirty, because a competitor’s page had quoted him one of those and he wanted a like for like comparison. I said neither, and that the truthful answer is a range with a caveat attached to it. He read that as evasion, which is fair, because every other page he had opened that week handed him a tidy number.

There is no tidy number. On a real carrier line there cannot be one, and the providers quoting minutes to two decimal places are either quoting an average as though it were a floor or selling something that is not a mobile line at all.

I run the farm these addresses come from. Real SIMs from Singapore carriers, modems on shelves in a rack, one port per modem. So the duration question lands on me a few times a week, and I have stopped giving the comfortable version of the answer.

The thing most people ask for is the thing that breaks their job

The default request when someone buys a mobile line is rotation. Fresh address every request, or every thirty seconds, or every minute. It gets asked for with total confidence.

Then I ask what the work is, and about half the time it is account work. Signing in, posting, managing listings, running a profile that has to be recognised as the same person tomorrow.

Rotation is the worst possible setting for that, and the reason sits in how a platform thinks. Anything with a login holds a running story about you: this credential was accepted here, on this device, and here is what it has done since. Your address is one line in that story.

When the address changes halfway through a signed in session, the platform sees a credential accepted at one place and the same session continuing from somewhere else, seconds later, with nothing in between that a human body could have done. Every fraud team on earth builds detection for exactly that shape, because it is what a stolen session cookie looks like in the wild.

So you get the challenge. A captcha, a verify email, sometimes a hold. And the usual conclusion is that there were not enough addresses in play, which leads to buying more of them and making the pattern louder.

The carrier owns the clock, not the seller

Here is the mechanical bit that explains why nobody can promise you a duration.

The address on a mobile line comes out of the carrier’s pool. It gets assigned when the modem attaches to the network, and it sits behind carrier grade address translation with a lot of other subscribers on that same network. It was never the seller’s to give.

What the seller controls is whether the modem reconnects. Rotation on a genuine mobile line is a reconnect with an API in front of it. The modem drops its data session, asks for another, and the carrier returns whatever it has free at that moment.

Everything past that belongs to the network operator. Whether you keep the same address across a reconnect. Whether the carrier moves you anyway during maintenance. What that next address was doing for the six months before it reached you.

I have seen a single address on an untouched line hold for most of a working day. I have also watched one move on its own inside the hour with nobody touching the modem. A signal dip and a reattach will do it. So will equipment somewhere in the network being swapped out at 3am by an engineer who has never heard of either of us.

What “sticky” means when you own the modems

The word carries more weight than it should, so it is worth splitting in half.

When you take a sticky port from me, the stable part is the port. Host, port number, credentials. That endpoint is yours and it stays put for as long as you are paying for the line.

The address behind that port is a separate matter and the answer belongs to the carrier. Sticky means I am not cycling that modem on a schedule, so it keeps whatever the network gave it for as long as the network is content to leave it alone.

The port is a promise I can make. The address is a promise I would be making on somebody else’s behalf, which is why I will not quote it in minutes any more.

A timer and a carrier change leave different fingerprints

This is the part I think gets conflated most, and it costs people accounts.

A natural carrier change is irregular. It fires when the network decides, at intervals that have nothing to do with your clock, and it happens to every handset on that carrier. The gaps are messy.

A forced rotation on a fixed timer is the opposite. Every ten minutes, exactly, all day. The intervals match each other because a scheduler produced them, and no real phone in Singapore has ever behaved that way. A phone gets a new address when it crosses between cells, when the network reshuffles, when its owner puts it on charge overnight and picks it up at seven. None of that lands on a grid.

So a strict timer stamps a machine shaped rhythm on top of an otherwise genuine carrier address. You paid for the realism and then wrote over it with a cron job.

The fix costs nothing. Rotate when a piece of work finishes rather than when a clock says so, and if you must keep a timer, let the interval wander a few minutes either side. That single change removes the one signal in your traffic that is unambiguously synthetic.

There is a second cost to a tight timer that people forget. A reconnect takes a few seconds during which nothing routes, and anything in flight dies with it. Rotating every thirty seconds means a real slice of your working hour is spent waiting for a radio to come back up.

Take the interval from the job

The number should come out of the work, not off a pricing page.

Start with your unit of work: the smallest chunk that has to finish on one address for the output to mean anything.

A marketplace seller account being warmed over two weeks has a unit measured in sessions. Sign in, do a few normal things, sign out, and that whole visit sits on one address. Whether the address is the same one tomorrow matters far less than whether it stayed still today.

A rank tracking sweep across a few hundred keywords is the opposite shape. No login, nothing to remember, every query independent. The unit is one request, and spreading those requests across as many addresses as the line will give you is pure upside.

A payment flow sits between the two. The unit runs from the first product page to the confirmation, and a cart that changes address in the middle is a fraud signal inside a system built specifically to hunt fraud signals.

Once you know the unit, the rule writes itself. Never change inside a unit. Do not be precious about changing between them. And if your unit is longer than the line reliably holds, the job needs breaking into smaller pieces, because no proxy fixes that for you.

You are choosing which wall to hit

Holding one address concentrates all of your volume onto it, and rate limits are counted per address by almost everything that counts anything.

So the ceiling on a long sticky session is arithmetic rather than detection. How many requests can one address make before the site starts keeping score.

Holding protects the session. Changing protects the volume. Neither is safer in the abstract, and the job in front of you decides which limit you reach first.

Two questions worth asking whoever sells you the line

Ask whether you can hold one address for the length of a session. A hard number of minutes with no qualification is a claim about a carrier the seller does not run, and it should make you suspicious. A range plus an admission that the network can override it is the answer from someone who has actually watched their own lines.

Ask whether you can force a change on demand. A real mobile line can be told to reconnect. Without that, you are stuck with the carrier’s timing, which is fine right up until the address you are sitting on turns out to be poisoned and you need off it now. The good answer includes a rate limit, because reconnecting a modem every few seconds is how a line gets flagged by the carrier itself.

There is a third one you get free. Ask what happens to the address when the modem reboots. “Nothing changes” means it is not a carrier line.

What I got wrong

I used to quote customers a sticky window in minutes, because they asked for one and it made the sale easier. It was an average wearing a promise, and it backfired in the most predictable way: every ordinary carrier reassignment then looked like our fault, and I spent support time explaining that the network had done a normal network thing.

I also ran my own scraping on a fixed five minute rotation for a long stretch, and blamed a climbing block rate on sites getting cleverer. Aligning the rotation to the job and adding a random offset fixed most of it. Same lines, same carriers, same code otherwise. I had been treating rotation as hygiene, something you run on a schedule the way you run backups, when it is part of the story your traffic tells.

If you want lines where you can hold an address for a full session and force a change when you decide to, that is what I run: real Singapore mobile IPs on actual Singtel, M1 and StarHub SIMs, sticky and rotating ports on the same line, and DNS that routes through the tunnel so you stop leaking your real resolver. There is a free trial of Singapore Mobile Proxy, and the code YT30 takes thirty percent off your first month.

Get new guides and videos first — join the Telegram channel.