BELT: Benchmark for Engine Limits & Throughput

[!NOTE] This was heavily inspired by abucnasty's work. I wanted to make a more universal, cross-platform version of the existing ps1 script.

BELT is a wrapper for the factorio --benchmark command, to make it more user friendly, more efficient to use, and to generate templated handlebars files with the gotten data.

Features

• Benchmarking - Benchmark a single save or a whole directory
• Cross-platform - Works on Windows, macOS, and Linux
• Multiple output formats - CSV, Markdown reports, and SVG charts
• Pattern matching - Filter save files by name patterns
• Async execution - Fast parallel processing
• Verbose metrics support - Per-tick charts and CSV exports for detailed Factorio metrics
• Sanitizer - Automatically parses and reports on belt-sanitizer mod output
• Smoothing and downsampling - Configurable smoothing for charts to handle large datasets

Quick Start

# Install BELT
cargo install belt

# Run benchmarks on all saves in a directory
belt benchmark ./saves --ticks 6000 --runs 5

# Filter saves by pattern and customize output directory
belt benchmark ./saves --pattern "inserter*" --output ./benchmark-results

Installation

Prerequisites

1. Factorio installed, BELT searches common installation paths, if none are found, please run with explicit --factorio-path.
2. Some save files to benchmark.
3. Rust if installing using cargo or building from source.

From Crates.io

cargo install belt

From GitHub Releases

1. Download the latest binary for your platform from Releases
2. Extract and place in your PATH

From Source

git clone https://github.com/florishafkenscheid/belt.git
cd belt
cargo install --path .

Usage

Basic Commands

# Basic benchmark with default settings
belt benchmark /path/to/saves

# Customize benchmark parameters
belt benchmark /path/to/saves --ticks 12000 --runs 10

# Filter saves and specify output location
belt benchmark /path/to/saves --pattern "benchmark*" --output /path/to/output/dir

Command Reference

belt analyze

*Arguments:

• <DATA_DIR> - The location of the csv(s) to generate charts based off of.
• <HEIGHT> - The height of the generated charts in pixels.
• <WIDTH> - The width of the generated charts in pixels.

Options:

belt benchmark

Arguments:

• <SAVES_DIR> - The location of the save(s) to be benchmarked.

Options:

belt sanitize

Arguments:

• <SAVES_DIR> - The location of the save(s) to be sanitized.

Options:

Global Options

Examples

Example 1: Basic Benchmarking

# Run a benchmark on the my-saves directory for 6000 ticks per run, and running each save file 3 times.
belt benchmark ./my-saves --ticks 6000 --runs 3

Example 2: Pattern Filtering

# Run a benchmark on the my-saves directory, only matching save files that start with "science" and outputting it to science-results/results.{csv,md}
belt benchmark ./my-saves --pattern science --output science-results

Example 3: Custom Factorio Path

# Run a benchmark on the my-saves directory, with an explicit path to the factorio binary
belt --factorio-path /path/to/factorio benchmark ./my-saves

Example 4: Specifying a mod list

# Run a benchmark on the my-saves directory and a mod directory
belt --factorio-path /path/to/factorio --mods-dir /path/to/mods benchmark ./my-saves

Advanced Usage

Best Practices

While belt benchmark offers sensible defaults, optimizing --ticks and --runs can refine your results. --ticks sets the simulation duration per run, while --runs determines the number of repetitions. Through testing, I've found that fewer runs with more ticks generally offers the most consistent UPS results for the shortest overall benchmark time, by reducing overhead from repeated Factorio launches. Experiment with these values for your specific saevs to find the optimal balance for accuracy and speed. However, for prolonged and thorough benchmarks, I recommend more runs in total, per save. This is because Factorio is deterministic, and when running BELT with verbose metrics, a "min" chart is generated. This chart is meant to combat any random noise that could slow down the Factorio benchmark, by only taking the fastest ticks of every run of a save.

Verbose Metrics

Here are all the verbose-metrics that are available: wholeUpdate,latencyUpdate,gameUpdate,planetsUpdate,controlBehaviorUpdate,transportLinesUpdate,electricHeatFluidCircuitUpdate,electricNetworkUpdate,heatNetworkUpdate,fluidFlowUpdate,entityUpdate,lightningUpdate,tileHeatingUpdate,particleUpdate,mapGenerator,mapGeneratorBasicTilesSupportCompute,mapGeneratorBasicTilesSupportApply,mapGeneratorCorrectedTilesPrepare,mapGeneratorCorrectedTilesCompute,mapGeneratorCorrectedTilesApply,mapGeneratorVariations,mapGeneratorEntitiesPrepare,mapGeneratorEntitiesCompute,mapGeneratorEntitiesApply,spacePlatforms,collectorNavMesh,collectorNavMeshPathfinding,collectorNavMeshRaycast,crcComputation,consistencyScraper,logisticManagerUpdate,constructionManagerUpdate,pathFinder,trains,trainPathFinder,commander,chartRefresh,luaGarbageIncremental,chartUpdate,scriptUpdate

Contributing

Any help is welcome. Whether you have never written a line of code, or simply don't know Rust. This is what the CI/CD pipeline is for! Bug reports and feature requests can be submitting through GitHub Issues.

If you want to contribute, please open an issue to discuss the proposed changes before submitting a pull request.

Standards

On every push a linter and formatter checks the code, so just write the code however you want and fix any errors that occur.

[!NOTE] To do this locally, run cargo fmt and cargo clippy -- -D warnings

I follow the Conventional Commits specification as a standard for my commit messages, I can only encourage you do the same.