Why is my game server lagging? Check these in order

"The server is lagging" describes at least four unrelated faults, and the fix for one makes no difference to the others. Worse, the most expensive response — buying a bigger plan — only helps with one of them.

Here is the order worth checking, cheapest first.

First: is it lag, or is it your connection?

Before anything else, establish whether the problem is server-side or client-side. They feel similar and have nothing in common.

• Client-side looks like: rubber-banding for you while other players are fine, high ping in the scoreboard, packet loss on a traceroute.
• Server-side looks like: everyone stuttering simultaneously, entities freezing, the world tick visibly slowing.

Ask in chat. If one person is struggling and five are not, stop reading and go look at their connection.

1. CPU steal time

Cost to check: two minutes. Cost to fix: a migration.

Steal time is the percentage of time your server wanted CPU and could not get it, because the physical machine was busy serving somebody else. On a Linux server:

top -bn1 | grep "Cpu(s)"

Look at the st figure. It should be zero. Consistently above two or three percent means the host has put more customers on the box than it can serve at peak, and no configuration change on your side will fix that.

This is first on the list because it is the one thing you genuinely cannot work around, and because it is common. It is also the reason we cap node allocation at 70% and publish that number.

2. Entity and object count

Cost to check: five minutes. Cost to fix: free.

Every game with building or item drops accumulates simulated objects, and the count only goes one direction. A server three weeks into a wipe is doing several times the work it did on day one, with the same hardware.

• Rust — entity.count in console. Also check decay settings; disabled decay means nothing is ever cleaned up.
• Minecraft — Spark profiler, or /tps first. Look for entity farms and unclaimed item drops.
• ARK — structure and tame counts per tribe; both have caps for a reason.
• Space Engineers — grid count, and whether concealment is running.

If your server was fine last month and is not now, and nothing else changed, this is almost certainly it.

3. Plugins and mods

Cost to check: ten minutes. Cost to fix: usually free.

One badly written plugin doing synchronous work inside a per-player or per-tick hook will cost you more than every other plugin combined. The good news is that every major framework has a profiler and they all point straight at the culprit.

• Minecraft — Spark. /spark profiler start, play for five minutes, /spark profiler stop. Sort by total time.
• Rust (Oxide) — plugin timing via the console.
• Garry's Mod — the Lua profiler.

Sort by total time, not by call count. A plugin called a million times cheaply is fine; a plugin called ten times expensively is not.

4. Memory pressure

Cost to check: two minutes. Cost to fix: sometimes a plan change.

Watch the memory graph in the panel over a few hours of real play, not immediately after a restart.

• Pinned near 100% with periodic freezes — genuinely short of memory, or leaking.
• Sitting around 60-70% — healthy; memory is not your problem.
• Sitting at 30% and still stuttering — memory is definitely not your problem. Go back to steps 1-3.

For Java games specifically, more memory is not automatically better: a larger heap means longer garbage collection pauses, and those pauses are what players feel. Our Minecraft RAM guide goes into that properly.

5. Only now, the hardware

If steal time is zero, entity counts are sane, no plugin is dominating the profile and memory is comfortable — then you have genuinely outgrown the plan, and upgrading will help.

That ordering matters because upgrading is the response every host is happy to sell you, and four times out of five it changes nothing. Any host that suggests it before asking what your entity count is has not looked.

The single-thread thing

Worth knowing as background: almost every game server is bottlenecked by one thread. Rust, Minecraft, Factorio, ARK, Valheim — they all have a main simulation loop that cannot be parallelised.

This is why core count barely matters and clock speed matters enormously, why a 5.7 GHz desktop-class part beats a 32-core server part for this workload, and why "we use enterprise Xeons" is a weaker claim than it sounds.

It is also why, past a certain point, no host can help you. A Factorio megabase below 60 UPS is below 60 UPS everywhere. At that point the fix is in the factory, not the invoice.

If you get stuck

Send us the profiler output. Support will read it with you, including if you are not a customer yet and are trying to work out whether moving would help. If the answer is that it would not, we will tell you that — it is a cheaper conversation than a refund.