# Performance benchmarks
LUBM (Lehigh University Benchmark) is the canonical OWL/SPARQL store
benchmark. We compare pgRDF against Apache Jena TDB and Apache AGE
at LUBM-10 (~1.3M triples) and LUBM-100 (~13M triples) scales.
## Layout (target)
tests/perf/
├── README.md (this file)
├── run-lubm.sh (driver; Phase 2)
├── queries/ (the 14 LUBM SPARQL queries)
└── data/ (generator output; gitignored)
## LUBM generator
LUBM ships a Java generator at https://swat.cse.lehigh.edu/projects/lubm/.
The runner script wraps it; the produced `*.owl` files are converted
to N-Triples and bulk-loaded via `pgrdf.load_file()`.
## Output
`target/perf-report.json` — a per-engine, per-query matrix of:
- cold-start latency (ms)
- warm latency (ms, median of 10 runs)
- result row count (sanity check)
Tracked release-over-release in [docs/09-release.md](../../docs/09-release.md).
## Gates per phase
- Phase 1: not run.
- Phase 2: LUBM-1 smoke (correctness only).
- Phase 3: LUBM-10 baseline numbers in the report.
- Phase 4: LUBM-100 comparison against Jena TDB + AGE.