Zarrzipan: A raster compression benchmarking tool

Zarrzipan is a compression benchmarking tool based on Zarr. It features a python API for declaratively describing numpy arrays and zarr compression pipelines to be benchmarked.

Note that this is experimental software written with significant LLM assistance. Some metrics such as memory usage and timing may not be accurately measured.

Usage

Zarrzipan includes a little CLI tool. Configure your arrays and jobs in a yaml file as demonstrated in zarrzipan.yaml and then run:

uv run zarrzipan compare | uv run zarrzipan render

The output of zarrzipan compare is ndjson, so if you are doing many runs you can build out your results over time:

uv run zarrzipan compare >> output.json

And render the output independently:

uv run zarrzipan render -f output.json

Benchmark Results

The following table summarizes the benchmark results described in zarrzipan.yaml:

Other tooling like this

geotiff-benchmark

The geotiff-benchmark tool is a gdal-based compression benchmarking solution targeting the GeoTIFF format specifically. While not know to us prior to the initial Zarrzipan development, geotiff-benchmark is similar in its goals, and allows testing a set of declaratively-configured GeoTIFF compression pipelines. Read the corresponding blog post to learn more about the tool.

What does "zarrzipan" mean?

Zarrzipan, pronounced like marzipan, is like "Zarr zip analysis", where zip is a general allusion to compression and not a reference to the specific zip archive format.