Lua ECS Library Benchmark
March 15, 2026 · View on GitHub
Overview
Benchmark suite measuring execution time and memory consumption of popular Lua ECS (Entity-Component-System) libraries. Built on luamark, a statistical benchmarking tool for Lua.
Tested Libraries
| Library | Version |
|---|---|
| Concord | 848652f |
| ecs-lua | c6be4a8 |
| evolved.lua | v1.10.0-0 (luarocks) |
| lovetoys | v0.4.0-2 (luarocks) |
| nata | eb9d49b |
| tiny-ecs | v1.3-3 (luarocks) |
Benchmark Tests
19 tests across 5 groups, each run at 3 entity scales (100, 1,000, 10,000):
| Group | Tests | Measures |
|---|---|---|
| Entity | 3 | Creation, destruction, allocation cost |
| Component | 4 | Field read/write, insertion, removal |
| Tag | 3 | Presence check, insertion, removal |
| System | 6 | Iteration throughput, scheduling, dispatch overhead |
| Structural Scaling | 3 | Structural ops with 20 registered systems |
Each test reports median execution time and memory consumption. See specifications for full test definitions and rationale for test selection reasoning.
Caveats: These are micro-benchmarks measuring framework overhead in isolation. They do not cover application-level concerns such as event messaging or serialization. Memory figures suit relative comparison, not absolute measurement.
Running the Benchmarks
Install:
- Lua 5.1+ (or LuaJIT) & Luarocks
- Python 3.12+ & uv: required to export the results.
- Taskfile: task runner
Run benchmarks with:
task
Libraries are automatically installed, provided the tools above are installed.
Results
- Frameworks with relative performance exceeding 20× the fastest are excluded from charts to improve readability.
- The results are expressed in terms of median execution time and memory usage; lower is better.
| Interpreter | Markdown | CSV |
|---|---|---|
| luajit-on | results.md | results.csv |
| luajit-off | results.md | results.csv |
| lua5.1 | results.md | results.csv |
Plots: results/plots/
Interpreting the results: results/README.md
Benchmark environment
OS: macOS 26.3.1
CPU: Apple M2 Max
Cores: 12 cores (12 threads)
Max Frequency: 3.50 GHz
Memory: 64 GB
Summary
Overall ranking using Borda count: each framework earns points based on its rank per test (last place = 1, first place = N). Higher score is better. Computed across all tests at 1,000 entities, grouped by category.
Plots
Detailed per-test plots: results/plots/
Contributing
Contributions, including adding new tests or ECS libraries, are welcome.