TripleTrail benchmarks

TripleTrail against Oxigraph, QLever, Jena, GraphDB and maplib on three standard SPARQL suites, and against PostGIS, QLever, Jena and GraphDB on its own spatial and routing benchmark. Every number comes from a run of the benchmark workflow on a rented server, with its inputs, logs and raw results in the repository.

How these were measured
Machine
Each tab names the machine, TripleTrail commit and date of its run, from provenance.csv. One engine at a time, all times at the client over HTTP.
Sparqloscope
The benchmark behind the QLever evaluation site: about 100 generated queries per dataset, each isolating one SPARQL feature. One cold run per query after an engine restart.
BSBM
Berlin SPARQL Benchmark with its own test driver: warm-up mixes, then timed query mixes from a single client. QMpH is query mixes per hour.
Trainmarks
DataTreehouse/trainmarks data and queries run over HTTP; its own site times libraries in-process, so its numbers are not directly comparable.
Spatial
TripleTrail's own bench v2: GeoSPARQL selections, nearest neighbours, spatial joins, aggregates, linear referencing, routing and OSM queries, the same questions asked of PostGIS/pgRouting in SQL. Warm figures are medians of several runs after one cold run.