// the find
ferrumc-rs/ferrumc
A reimplementation of the minecraft server in rust.
A from-scratch Minecraft 1.21.8 server in Rust, built on Bevy ECS and Tokio. It targets the average person running a server for friends, not plugin-heavy networks — the pitch is genuine multi-core utilization and low memory use instead of Paper's single-threaded tick loop.
1. Bevy ECS as the concurrency model is the right call here — systems run in parallel by default, no manual thread management, and you get lockless access patterns that a Java reimplementation could never match. 2. They wrote their own NBT and Anvil libraries using experimental APIs and raw SIMD rather than wrapping existing crates that were too slow — that's exactly the kind of decision that separates a serious performance project from a hobby port. 3. Tracy profiling is a first-class build feature (`cargo run --features tracy`), not an afterthought — you can actually measure where time goes instead of guessing. 4. Existing vanilla worlds import via Anvil format, so this isn't a fresh-start-only proposition.
1. Requires nightly Rust — `rust-toolchain.toml` pins the version which helps, but you're one upstream API churn away from a broken build, and this will be a permanent tax unless they eventually stabilize their feature usage. 2. The Java plugin ecosystem is a hard wall — no Bukkit/Spigot/Forge/Fabric compatibility, so anyone with existing plugins or mods is starting over. The plugin system that will replace this doesn't exist yet. 3. Entities and PvE are barely started — pigs exist, mobs are listed as 'upcoming', which means the server is currently closer to a creative-mode sandbox than something a vanilla survival group would run. 4. Terrain generation intentionally deviates from vanilla when accuracy would hurt performance — reasonable tradeoff in principle, but the seams where worldgen diverges from expected vanilla behavior will produce confused bug reports for years.