The repository contains standalone performance and utility benchmarks for stochastic aggregation and differential privacy. The exact paper environment, experiment matrix, bounds, result columns, and plotting commands are documented in repro.md.
Run commands from the repository root. Build the paper executables with:
cmake --build build/release --target \
as_tpch_benchmark as_clickbench_benchmark \
dp_benchmark_runner| Workload | Executable | Configuration |
|---|---|---|
| TPC-H AS | as_tpch_benchmark |
benchmark/configs/utility/as_tpch_sf30.json |
TPC-H variable m |
as_tpch_benchmark |
benchmark/configs/utility/as_tpch_sf30_m_sweep.json |
| ClickBench AS | as_clickbench_benchmark |
benchmark/configs/utility/as_clickbench_graviton.json |
| TPC-H DP utility | dp_benchmark_runner |
benchmark/configs/utility/dp_tpch_sf30_utility.json |
| TPC-H Q01 AVG | dp_benchmark_runner |
benchmark/configs/utility/dp_tpch_sf30_q01_avg_*.json |
The runners accept --config <path> --dry-run to validate a JSON configuration without opening a database. Generated
databases and results are ignored by Git and written below benchmark/results/ unless a config selects another path.
- Aggregation microbenchmarks compare aggregate kernel implementations.
- IMDb/JOB exercises person-level privacy over a multi-table workload.
- SQLStorm provides optional query-coverage smoke tests.
- Utility evaluation documents DP bounds, metrics, and diagnostic tools.