How to Reduce Lag on Your Minecraft Server

Understanding TPS

TPS is ticks per second, and it is the heartbeat of the server. Everything the world does, mob movement, block updates, redstone, growth, happens once per tick, and the target is 20 of them a second. That gives each tick 50 milliseconds to finish in. Go over and the next tick starts late, TPS falls, and the whole world slows down together: this is why lag makes furnaces smelt slower rather than making the game look choppy. Type /tps in the console to see it. Paper reports three averages, over one, five and fifteen minutes, and the one-minute figure is the one that answers "is it bad right now".

What are your own numbers?

Everything below only helps once you know which number is out of line on your server. You do not have to work that out by hand: the AI performance optimizer in your panel measures your own server in about seven seconds, from tick time and tick rate to entities per type, chunks, memory and your plugins, and names the culprit in one sentence with the number behind it.

Three Types of Lag

Three different problems get called lag, and only one of them is yours to fix. Telling them apart first saves you from optimising a server that was never the slow part.

Pre-Generate Your Chunks

Generating terrain is the most expensive thing a server does. Biomes, caves, structures, ore and lighting all get worked out the first time somebody walks that way, and four players riding horses in four directions can drop a healthy server to single-digit TPS on their own. Doing it in advance, while nobody is on, moves that cost to a moment when it does not matter.

1. Install Chunky

   It is an ordinary plugin jar in the plugins folder, and it works on any of our Bukkit-based server software.

2. Set the radius you actually want

   Pick the same number as your world border, in blocks. Generating far past the border is hours of CPU for terrain nobody will ever stand on.

   /chunky radius 5000

3. Start it on an empty server

   The generation itself is heavy, so run it overnight or before you hand out the address. It resumes if it gets interrupted.

   /chunky start

Lower Your View Distance and Simulation Distance

View distance is a radius, so the cost of raising it is square rather than linear. Our own panel takes 2 to 32, and across that range one player goes from 25 chunks to 4,225. Multiply by the number of people online and you can see why this is the first dial to touch. Dropping from 10 to 8 costs almost nothing anyone notices and takes about a third of the chunks off the server.

Simulation distance is the one that pays. View distance decides what a player can see; simulation distance decides how far the world keeps running, so it is what actually costs ticks. Setting it a few chunks below view distance is the usual answer. These are the 5 performance settings our own panel exposes, with the text we put next to them.

Clean Up Entities

Every mob, dropped item, item frame, armour stand and minecart is an entity, and the server walks the whole list every tick. Five hundred chickens packed into ten blocks cost more than fifty players walking around. Paper can thin this out without anyone noticing: activation ranges and tick rates in paper-world-defaults.yml make distant entities tick less often, and spawn limits in bukkit.yml cap how many can exist at once. Item frames and armour stands in a build are the sneaky ones, because nothing kills them and they never move.

Optimize Redstone

A redstone update cascades: one change fires block updates through everything connected to it, and a fast clock does that hundreds of times a second. Paper ships an alternate redstone implementation that is measurably cheaper for large circuits and is worth turning on before you start banning designs. Hoppers deserve their own rule: each one checks for items every tick whether anything is moving or not, so a two-hundred-hopper sorting system runs all day at full price.

Use Spark to Find the Bottleneck

Everything above is a guess until you measure. Spark is a profiler that tells you where the ticks went: entity ticking, chunk loading, one plugin's event handler, or the garbage collector. Start it while the server is actually struggling, leave it running a few minutes, stop it, and it hands you a link with the breakdown. Paper 1.21 and up ships with it already installed, so on most of our servers there is nothing to install first.

Plugin-Related Lag

Plugin count is not the number that matters. Fifteen well-chosen plugins beat forty where half overlap, but one plugin scanning chunks on the main thread beats them all for damage. If a Spark profile puts a single plugin at the top, look for a newer version, then for a config option that turns off the expensive part, and then for something else that does the job. Going through the list twice a year and deleting what nobody uses costs an hour and pays for itself.

When to Upgrade Your Plan

If you have pre-generated your chunks, cut simulation distance, thinned the entities and profiled with Spark, and the one-minute average still will not hold 20, then the machine is the limit and not the settings. Sitting above eighty percent memory all day, or CPU spikes during ordinary play, is the signal. All 7 of our plans list what they give you, and our memory guide has which of them fits which kind of server.

What each plan gives you, straight from the price list.

Plan	vCPU	Memory	Storage
Mini	2 vCPU	4 GB	60 GB SSD
Starter	2 vCPU	8 GB	60 GB SSD
Standard	3 vCPU	12 GB	75 GB SSD
Premium	4 vCPU	24 GB	150 GB SSD
Ultimate	6 vCPU	48 GB	300 GB SSD
Enterprise	8 vCPU	64 GB	375 GB SSD
Dedicated	9 vCPU	96 GB	450 GB SSD

What comes with every plan, on every tier, is on the full feature list.

Questions people ask when the server drags

When the tuning is finished and it is still slow

Pre-generate the chunks, drop the view distance, thin out the entities, and measure again. If the tick time still sits near fifty with memory to spare, none of the four steps above is the answer any more: one core is out of time. That is the case the Performance line is for, on an AMD Ryzen 9 5900X, with the same memory you have now.

Starter Performance, 8 GBStandard Performance, 12 GBPremium Performance, 24 GB

How TPS and MSPT are readA dedicated thread against a shared vCPU