Reading a profile instead of guessing
An hour of learning, once, and lag stops being mysterious. The three shapes every profile falls into, and what percentages do not mean.
Everything published under this tag, newest first.
An hour of learning, once, and lag stops being mysterious. The three shapes every profile falls into, and what percentages do not mean.
They cannot see each other and they are competing. CPU contention, the memory arithmetic, and the port map nobody writes down.
The useful question is not whether to restart. It is whether yours is buying headroom or hiding a leak that is getting worse.
Terrain generation is the most expensive thing a server does. Move it to a time nobody is playing, and set the edge before you start.
Three more files hold the settings that decide whether a busy server runs well. Which entries matter, and which ones hide a problem.
Everyone pastes the same twenty arguments and nobody knows which one to change. What each part of Aikar's flags actually asks G1 to do.
Sixty entries, six that matter. Which lines change what players feel, which cost CPU, and which are folklore repeated from old forum posts.
Most stutter blamed on the CPU is a scheduled write. How to tell whether the disk is your problem, and what faster storage does and does not buy you.
A server that is late rather than failing leaves no trace in the logs. How to establish the fact, read a profile, and name the guilty plugin in a minute.
Eight causes, ranked by how often they are the answer. Hardware is last, and the profiler in step two resolves most servers before you get there.
Giving a Minecraft server more memory usually makes it worse. Why the heap size is not a speed setting, and the numbers that work.
A game server simulates its world on one thread. The most expensive misunderstanding in hosting, and the measurement that settles it in a minute.