GSTools-Benchmarks — published performance history

This repository contains the published benchmark history for GSTools. It is generated and updated automatically by GSTools' CI

Do not edit this repository by hand. Its contents (results, report, landing page) are overwritten by the GSTools publish workflow on every accepted change. Manual commits will be force-pushed over.

What this is

GSTools tracks the runtime and peak memory of representative two-point statistics workflows — variogram estimation, global kriging, and (conditioned) random field generation — on the GSTools-Core (Rust) backend, across thread counts. This repo stores one measurement per commit merged to GSTools main, so the published site shows how performance evolves over the project's history.

Measurements are produced with Airspeed Velocity (ASV). For what each case does and how to run the benchmarks yourself, see the benchmark guide in the main repository: benchmarks/README.md.

How it is updated

The asv-benchmark-publish.yml workflow in GeoStat-Framework/GSTools runs on every push to main that touches benchmark-relevant paths. For each accepted commit it:

1. restores the cumulative results already stored here,
2. benchmarks the newly accepted commit on a fixed CI machine (github-actions-ubuntu),
3. appends the new result and rebuilds the report, then
4. commits and pushes the updated history back to this repository.

Pull-request benchmarks are not stored here — those run per-PR and are only posted as PR comments. This history contains one clean data point per merged commit.

Layout

index.html     redirect to asv/index.html (the landing page)
asv/           native ASV HTML report — interactive, per-commit history
results/       raw ASV result JSON — the source of truth
.nojekyll      disables Jekyll so the static site is served as-is

Reading the results

• Compare performance commit-over-commit on the same CI machine. Absolute times are not comparable across different machines, so treat the numbers as relative trends, not hardware benchmarks.
• The PR comparison uses ASV's Mann-Whitney U test at a 5% threshold; a ~ marker means the ratio moved but the change was not statistically significant on the (shared, noisy) runner.
• time_* benchmarks measure runtime; peakmem_* measure whole-process peak memory.

Links

• GSTools: https://github.com/GeoStat-Framework/GSTools
• Benchmark guide: https://github.com/GeoStat-Framework/GSTools/blob/main/benchmarks/README.md
• Airspeed Velocity: https://asv.readthedocs.io