mc-bench — Minecraft Server Load Benchmark
Benchmarks vanilla vs patched Minecraft server-26.1 under simulated load.
Setup
# JDK 25 required
JAVA=/tmp/jdk25/bin/java
# Server jar (bundler)
SERVER_JAR=/home/fox/Downloads/server.jar
# Install both servers
bash bench-server.sh vanilla 40
bash bench-server.sh patched 40
Tools
| File |
Purpose |
bench-server.sh |
Full benchmark: startup + /reload + bot wave |
gen-modpack-tags.py |
Generates synthetic depth-N diamond tag data |
raw-connect.mjs |
Raw TCP handshake bot (version-agnostic) |
sim-players.mjs |
mineflayer bot runner (requires matching protocol version) |
rcon.py |
RCON command injection |
What it measures
- Startup time (wall clock, JVM launch → "Done")
- /reload time (RCON-triggered, measures tag reload)
- Connection capacity (how many handshakes the server accepts simultaneously)
- Handshake latency (p50/p95/max)
Notes
- Real-world improvement requires a genuine modpack with tag depth ≥ 10.
- Synthetic tag data uses HashMap iteration (hash-ordered), which may not trigger
worst-case DFS. Real modpacks load tags alphabetically (base before root) which
reliably triggers the exponential in vanilla.
- See
tests/bench/LoadSimBenchmark.java for controlled algorithmic benchmark.
