RAM: the number that everyone gets wrong
The most common hosting mistake is buying too much RAM. Minecraft's server loop is mostly single-threaded; a 16 GB allocation on a weak CPU performs worse than 4 GB on a strong one. Oversized heaps make Java's garbage collector pause longer, which players feel as periodic stutter.
| Setup | Players | Sensible RAM |
|---|---|---|
| Vanilla / Paper, no plugins | 2–5 | 3 GB |
| Paper + plugins | 5–15 | 4–6 GB |
| Paper + many plugins, big view distance | 15–30 | 6–8 GB |
| Fabric modpack (light) | 3–8 | 6–8 GB |
| Fabric/Forge modpack (heavy) | 3–10 | 8–12 GB |
-Xms4G -Xmx4G with G1GC. Keeping minimum and maximum equal avoids resize pauses. More than 10–12 GB for a single server almost never helps.Choosing the server software
Paper
The default answer for 90% of servers. A Spigot fork with hundreds of bug fixes, better entity handling and plugin support. If you cannot name a reason to avoid it, use it.
Spigot
Use only when a specific plugin refuses to run on Paper. Otherwise it is strictly a downgrade.
Fabric / Forge
For modpacks. Fabric is lighter and updates faster; Forge has the bigger legacy catalogue. Do not mix plugins into modded servers — pick one ecosystem per server.
Vanilla
Pure Mojang jar. Choose it for technical farm servers or exact vanilla mechanics; everything else performs worse on it.
TPS: reading your server's pulse
TPS (ticks per second) is the server's heartbeat, targeting 20. On Paper, run /tps in console: 19–20 is healthy, 17–19 is "players notice", below 15 means something is structurally wrong. The second number to watch is /timings on (or spark profiler) output — it names the exact plugin, entity type or chunk region eating your tick.
The usual TPS killers, in order of how often they are guilty: runaway mob farms with thousands of stacked entities, hopper lines moving items across long distances, map-art canvases with thousands of item frames, chunk-loaders keeping hundreds of chunks ticking, and one player's flying-machine redstone contraption. Each is fixable with entity culling, hoppers swapped for water streams, and plugin limits.
A starter plugin shelf
Do not install thirty plugins on day one. Each one is a dependency, an update burden and a potential lag source. A stable starter set: EssentialsX (homes, warps, basic commands), LuckPerms (permissions), Chunky (pre-generate the world so players never wait on chunk gen), Vault (economy bridge), and a backup plugin such as ArmoredElytra-style no — keep it simple: a cron job or plugin that zips the world folder on a schedule. Everything else can wait until a real need appears.
Backups: the least glamorous lifesaver
Rule: a backup you have not tested restoring is not a backup. Zip the world folder, verify the archive opens, and keep at least three generations — daily, weekly, and one "before the big update" snapshot. World corruption from crashes is rare; corrupted world folders from failed manual edits are not. Store backups off the server machine if you can, because a disk failure takes the copy with the original.
Onboarding players, or why servers stay empty
Technical stability is half the job. The other half: a spawn area that explains itself in ten seconds, a short welcome message with the three commands a newcomer actually needs, and rules visible in-game, not on a Discord nobody joined. Servers that grow share one trait — a returning player can figure out "what's new" without asking anyone. A small, regularly updated notice board at spawn outperforms any amount of advertising.