sorting_simulation

Sorting Simulation

Sorting simulation to quantify the how the different techniques used in DuckDB's sorting implementation contributed to its performance.

Simulation

Run make for an optimized build, or make debug for a debug build. Parameters can be changed in the main() function at the bottom of the simulation.cpp script. The current parameters can are set to take up less than 16GB of memory. The script is run with ./simulation, which generates CSV files.

Cache misses and branch mispredictions are counted with macOS Instruments. Run make trace and run python3 trace.py.

Analysis

Plots were created in a Jupyter notebook. The relevant Python packages can be installed with:

python3 -m pip install notebook matplotlib seaborn duckdb pandas --user

Run jupyter-notebook (or jupyter notebook depending on the OS), then select the analysis.ipynb notebook and run all cells to create the plots.