Guides6 min read
How to choose Minecraft server hosting for your community
Choose Minecraft hosting by server software, workload, location, and recovery options. A practical checklist for a first SMP, plugins, or a modpack.
Fix Minecraft server lag by identifying whether the delay comes from simulation ticks, network delivery or the player’s client. On Paper, capture a spark profile during the problem and change the work the profile identifies. A copied optimization configuration can alter gameplay without addressing the actual bottleneck.
Which kind of lag is happening?
| Observation | Check first |
|---|---|
| World simulation slows for many players | Server TPS/MSPT and a busy-period profile |
| One player has delayed actions but ticks are healthy | That client and its network path |
| Low frame rate while server metrics remain healthy | Client rendering, mods and local resources |
| Brief pauses during travel or saves | Chunk generation, I/O, garbage collection and scheduled work |
| Server disappears or restarts | Crash logs, process limits and host events |
Read TPS and MSPT together. At the conventional 20 TPS target, sustained tick work must fit the normal 50 ms budget, but an average can hide spikes. Modified tick rates and region-threaded software need their own interpretation.
1. Capture the problem with spark
Paper includes spark on supported modern releases, beginning with Minecraft 1.21. Verify your build before adding another profiler. Record game/server/Java versions, extensions, player count, activity and effective distance settings.
/spark tps
/spark profiler start --timeout 120These are in-game administrator commands; use the console’s expected command format if running there. Capture representative activity during the two-minute example interval. Keep a healthy comparison sample and review reports for sensitive information before sharing them.
2. Follow the expensive work
| Profile or correlated evidence | Candidate investigation |
|---|---|
| Extension callbacks dominate | Resource version, configuration and unnecessary repeated work |
| Entity processing dominates | Entity counts, active areas and the gameplay requiring them |
| Chunk generation dominates during exploration | Travel workload, generation settings and planned pregeneration |
| Save/I/O pauses align with backups | Storage latency and competing operations |
| Memory pressure or long collection pauses | Heap, total limit and retained growth |
A high percentage in a sampled call tree identifies where observed execution time accumulated; read the surrounding work and sampling context before naming a cause. Waiting, asynchronous work and a symptom outside the captured interval may need additional evidence.
3. Optimize Paper without breaking the world’s rules
- Back up the exact configuration and record a baseline.
- Select one change justified by the profile, using documentation for the installed release.
- Check inherited defaults and per-world overrides rather than assuming one file controls every world.
- Repeat similar gameplay and compare tick-time distribution, errors and affected mechanics.
- Keep the change only if its benefit meets the community’s goals; otherwise restore the previous value.
Start with view versus simulation distance if loaded or simulated areas are the identified work. Entity activation, spawning and other world settings can change farms or behavior; avoid treating the lowest possible values as a universally correct preset.
Pregeneration is a workload decision
Preparing a bounded area can move some generation work out of active play, but it consumes disk and CPU and does not eliminate every future loading task. Use a compatible tool, a defined boundary and a maintenance plan. Observe free space and performance while it runs rather than launching an unlimited job beside peak gameplay.
4. Upgrade only the constraint that remains
If required tick work remains too slow, use the CPU guide. If memory failures are present, use OOM diagnosis. Adding heap will not automatically fix an expensive serial task, and more cores will not automatically shorten it.
Keep before/after evidence and the final configuration with the deployment record. Explore the Minecraft topic hub for supporting guides and compare Minecraft hosting when the measured workload needs different resources. No performance gain is claimed here without a workload-specific test.
