Measure the simulation loop
Player count is only one input. Track tick duration, CPU saturation on the busiest thread, memory growth, save time, packet loss, and latency distribution during realistic matches. Mods can move the limit long before the advertised slot count.
Choose a region from player latency rather than operator convenience. Test the actual route from expected player locations and keep DDoS response ownership explicit.
- Tick duration
- Player latency distribution
- Save and restart time
- Mod and version inventory
Make updates reversible
Pin game and mod versions, stage updates with a representative save, and keep a known-good artifact. Announce restart windows from measured save and startup times.
Backups must include configuration, permissions, world state, and any external database. A copied directory is not enough until it has produced a working replacement server.
Run a full match under load, force a restart, restore the latest save on a clean server, and compare tick and latency measurements.