Featured suites
Recognised public benchmark suites, latest commit. The sparq
column is this run; competitor columns (QLever / Oxigraph / EYE / RDFox,
with pinned versions) are versioned static data gathered once — not measured per
commit — from bench/dashboard/competitors.json. A cell shows n/a
where no competitor number was gathered at this scale (no fabricated numbers); real cited
numbers at their own scale/machine appear as external reference baselines beneath
each suite. A tier badge shows whether these sparq numbers come from the
weekly EC2/nightly heavy-benchmark series (preferred when present) or the noisier
per-commit gh-runner band.
Same-box SPARQL comparison
A self-consistent head-to-head: sparq and
Oxigraph (plus QLever when the best-effort Docker server
comes up) run on one ephemeral EC2 box, over the same generated featured
corpus and the same query files, min-of-N in COUNT mode, with solution counts
cross-checked. Absolute µs are this box only (a 4-vCPU Graviton, not the
gh-runner CI band) — read them for cross-engine ordering, not as a cross-machine
reference. n/a where an engine produced no timing (e.g. QLever skipped).
Benchmark families
Every query type sparq benchmarks, grouped into capability families. The table
shows each metric's latest value (smaller is better —
customSmallerIsBetter) with a Δ-vs-best pill. Expand Trends in a family
to load its history charts (loaded on demand to keep the page fast). Hover any label or the
raw stem_us beneath it for the dataset + query it measures.
Scaling
How a metric scales with dataset size / depth, for suites run at multiple sizes (e.g. Deep Taxonomy depths, WatDiv scale factors). The size/depth axis is read from the metric name.