Python HPC

Incremental N-Body optimizations (all-pairs) in order to evaluate and compare the performance of Python translators in the HPC environment. This work has served as the basis for the following publications:

• Springer - Performance Comparison of Python Translators for a Multi-threaded CPU-bound Application

@InProceedings{10.1007/978-3-031-05903-2_2,
  author="Milla, Andr{\'e}s
  and Rucci, Enzo",
  editor="Pesado, Patricia
  and Gil, Gustavo",
  title="Performance Comparison of Python Translators for a Multi-threaded CPU-Bound Application",
  booktitle="Computer Science -- CACIC 2021",
  year="2022",
  publisher="Springer International Publishing",
  address="Cham",
  pages="21--38",
  abstract="Currently, Python is one of the most widely used languages in various application areas. However, it has limitations when it comes to optimizing and parallelizing applications due to the nature of its official CPython interpreter, especially for CPU-bound applications. To solve this problem, several alternative translators have emerged, each with a different approach and its own cost-performance ratio. Due to the absence of comparative studies, we have carried out a performance comparison of these translators using N-Body as a case study (a well-known problem with high computational demand). The results obtained show that CPython and PyPy presented poor performance due to their limitations when it comes to parallelizing algorithms; while Numba and Cython achieved significantly higher performance, proving to be viable options to speed up numerical algorithms.",
  isbn="978-3-031-05903-2"
}

• Bachelor thesis - Un Estudio Comparativo entre Traductores de Python para Aplicaciones Paralelas de Memoria Compartida
  • Presentación

@phdthesis{milla2022estudio,
  title={Un estudio comparativo entre traductores de Python para aplicaciones paralelas de memoria compartida},
  author={Milla, Andr{\'e}s},
  year={2022},
  school={Universidad Nacional de La Plata}
}

• PyConAr 2021 - Acelerando aplicaciones paralelas en Python: Numba vs. Cython

@misc{milla_rucci_pycon,
  address = {PyCon 2021},
  type = {Conferencia},
  title = {Acelerando aplicaciones paralelas en {Python}: {Numba} vs. {Cython}},
  url = {https://eventos.python.org.ar/events/pyconar2021/activity/448/},
  author={Milla, Andr{\'e}s and Rucci, Enzo},
  month = oct,
  year = {2021},
}

• CACIC 2021 - Acelerando código científico en Python usando Numba

@inproceedings{milla_rucci_cacic,
  title={Acelerando c{\'o}digo cient{\'\i}fico en Python usando Numba},
  author={Milla, Andr{\'e}s and Rucci, Enzo},
  booktitle={XXVII Congreso Argentino de Ciencias de la Computaci{\'o}n (CACIC 2021)},
  year={2021}
}

• CACIC 2022 - Comparación de Rendimiento y Esfuerzo de Programación entre Numba y Cython para una Aplicación Multi-hilada de Alto Rendimiento

@inproceedings{milla2023comparacion,
  title={Comparaci{\'o}n de Rendimiento y Esfuerzo de Programaci{\'o}n entre Numba y Cython para una Aplicaci{\'o}n Multi-hilada de Alto Rendimiento},
  author={Milla, Andr{\'e}s and Rucci, Enzo},
  booktitle={XXVIII Congreso Argentino de Ciencias de la Computaci{\'o}n (CACIC)(La Rioja, 3 al 6 de octubre de 2022)},
  year={2023}
}

Structure

The source code is located in the src directory, which contains the following subdirectories:

• versions: Contains the source code for each tested version.
• benchmarker: Script to run the benchmarks.
• test: Tests of the developed versions.
• core: Utils common to the modules.

Contribution

Requirements

Package	Version
Python	3.8.10
PyPy	7.3.1
Cython	0.29.22
Pip	21.0.1
Virtualenv	20.0.17

In Debian-based environments they can be installed with the following command:

apt-get install python3 python3-pip cython pypy3 virtualenv

Execution

1. Install the dependencies:

   make install

2. Configure the benchmark parameters in the config.toml file and run:

   make benchmark

The results will be visible in the benchmarks directory.

• Note: If you don't have the ICC compiler, you can choose another through the Makefile.

Contact

• [email protected]
• [email protected]