If you have ever run a game server that was smooth all afternoon and then stuttered the moment your evening crowd logged in, you have probably met an oversold host. This post explains what overselling actually is, the signs that give it away, and the questions you can ask before you hand over any money. The aim is to make you a harder buyer to fool, whether or not you ever host with us.

What overselling actually is

Overselling is selling more capacity than the hardware physically has, on the bet that not everyone uses their share at the same time. A box with 128GB of RAM and 16 cores gets sold as if it had far more, because most servers sit idle most of the time. On a quiet Tuesday that bet pays off and everything feels fine.

The bet fails at peak. When enough tenants get busy together, the box runs out of real cores and real memory, and the operating system starts sharing what it does not have. For a game server that shows up as tick jitter: the main loop misses its deadline, TPS drops, and nothing in your own logs explains why, because the cause is a neighbour you cannot see.

In fairness, overselling is not fraud, and a lightly oversubscribed box can run well for a long time. It is a business model, and the margin comes from the gap between what is sold and what is installed. The problem for you is that the gap only shows up when you least want it to.

The signs to look for

A few things tend to travel with an oversold plan. None is proof on its own, but together they tell a story.

  • A price the hardware cannot cover. Work out the rough cost of the RAM and CPU you are being promised, add power, bandwidth and a data-centre rack, and see whether the monthly price could plausibly pay for a fair share of it. If a plan promises 8GB of reserved memory for the price of a coffee, the memory is not really reserved.
  • Vague specs. "Premium CPU", "high-performance cores", "enterprise hardware" with no model number. A host that has bought good hardware tends to name it, because the name is the selling point. Missing numbers usually means the number is not flattering.
  • "Burstable" and "up to" language. Burstable RAM means you can use the headline figure only while the neighbours are quiet. It is oversubscription written into the plan. The same goes for "up to" core counts and CPU percentages that read like a share of something shared.
  • Lag that arrives on a schedule. If your server is fine off-peak and drops TPS every evening or every weekend, and your own player count and plugins do not explain it, the likely cause is contention on the host at the busy hours.

How to test before you buy

You do not have to guess. You can ask, and you can measure.

Ask the host, plainly, whether cores and RAM are reserved or shared, and whether they oversell. A straight answer is a good sign in itself. Ask to see live metrics rather than a marketing screenshot: a public Grafana or status dashboard tells you what the fleet is actually doing right now, not what it did on a good day.

If you can get a trial or a monthly plan, run your own test. Load a real world, invite a realistic number of players, and watch TPS and MSPT during the busiest hour you can arrange. A box with genuine headroom holds its tick under load. An oversold one degrades exactly when the players arrive. Our own Minecraft lag checklist walks through reading those numbers if they are new to you.

One more question worth asking: what happens when a container briefly spikes over its memory? On a properly reserved plan a short spike is absorbed. On an oversold box it can mean an out-of-memory kill, because the host never had the spare memory to lend.

How we handle it, stated plainly

We do not oversell, and the whole of our stack rests on that one rule. A core we allocate to you is yours, and the RAM we promise is reserved rather than borrowed from a quiet neighbour. That is the boring, expensive part of the business, and it is the part that makes the tuning worth doing, because you cannot tune your way out of a box carrying three times too many tenants. There is more detail in our full stack write-up.

Because the claim is easy to make and hard to prove, we show the work instead. Our resources are public and live on our Grafana and status dashboards, so you can watch the fleet running rather than take our word for it. If you are weighing up whether a shared plan is even the right shape for you, our guide on dedicated vs shared game hosting covers that decision honestly.

The point of all this is not to make you suspicious of every cheap plan. Plenty of small servers run happily on shared hosting for years. The point is that you can now tell the difference between a plan that is cheap because it is efficient and one that is cheap because it is oversold, and you can ask the two or three questions that separate them.

Browse game hosting