Specialised Query Executor for TPC-H Q18

Prerequisites

• Python 3.8+
• duckdb
• pandas
• pyarrow

Install dependencies with:

pip install -r requirements.txt

Directory Structure

project_root/
├── benchmarks/            # Benchmark strategies
├── data/                  # TPC-H tables in Parquet format
├── engine/
│   ├── duckdb_engine.py
│   └── custom_engine.py
├── results/
│   ├── benchmark/         # Custom engine query results (CSV)
│   └── target/            # DuckDB query results (CSV)
├── init.py                # Data generation script
├── main.py                # Main entry point
└── summary.csv

1. Initialize Project (Generate TPC-H Data)

Run the init command to generate TPCH tables in Parquet format and initialize result directories:

python main.py init

The following files will be created for scale factors [0.5, 1, 2, 5]

data/sf{0.5, 1, 2, 5}/
├── customer.parquet
├── lineitem.parquet
├── nation.parquet
├── orders.parquet
├── part.parquet
├── partsupp.parquet
├── region.parquet
└── supplier.parquet

2. Run the Benchmark

python main.py benchmark

Optional arguments:

• --out summary.csv: Output file for the benchmark results (default: summary.csv)
• --benchmark 5: Number of timed repetitions per scale factor after one warm-up run (default: 5)
• --strategy <strategy>: Benchmark execution strategy:
  • interweave (default)
  • duckdb_first
  • custom_engine_first
• --enable_profiling: Enable detailed profiling for the custom engine

This will:

• Clear previous results in results/benchmark and results/target
• Run both engines for scale factors 0.5, 1, 2, 5
• Run one warm-up iteration first, then average the next --benchmark timed runs
• Save query results as CSVs in the results folders
• Write benchmark results to summary.csv

3. Data Correctness Check

python main.py check

This will compare all matching CSV files in results/benchmark and results/target and print any mismatches.