Motivation

Python is the most popular programming language

Motivation

Python is the most popular programming language

Motivation

Ever more production systems (web-applications/-services) are written in Python

Motivation

CPython developers focus on performance improvements
No current Python Enhancement Proposal (PEP) is about increasing the sustainability (in terms of energy efficiency) of CPython
No knowledge about how performance improvements impact energy consumption of CPython
No experiments reported that investigate energy consumption of programs across interpreter/compiler versions

The Problem

Source: Pereira, R. et al.
“Ranking programming languages by energy efficiency” (2021)

The Problem

Source: Georgiou, S. et al. “Analyzing programming languages’ energy consumption: An empirical study” (2017)

Experiment Design

MiniTwit a Twitter-like micro-blogging application. Originally, Armin Ronacher’s example application for the Python web-framework Flask.

Experiment Design

Results

Impact for Practitioners

For developers and maintainers, we recommend to switch to CPython 3.12 the currently the most energy efficient CPython interpreter
Energy consumption drops by more than 8% compared to the most energy consuming CPython 3.10.
For the developers of CPython, the performance improvements implemented in recent versions of CPython render CPython more sustainable.
We suggest to execute the Python Performance Benchmark Suite in a setup similar to ours to continuously monitor also energy efficiency of future CPython versions.

Impact for Researchers:

Relevant for future language improvements are questions like

a) Which implementation details cause a Python interpreter to draw more power or to become faster and thereby impact energy consumption?
b) What is the precise reason for that power draw increases for certain versions of CPython that are optimized for performance?

I hope that this work inspires others to investigate such questions and thereby help to increase sustainability of the reference implementation of the currently most popular programming language.

A replication kit is available on GitHub: https://github.com/HelgeCPH/cpython_energy

Future Work

Extend this experiment by executing the Python Performance Benchmark Suite to investigate more thoroughly the relation between performance improvements and power draw.
Conduct these on more than one model of the Raspberry Pi but on various newer models and other architectures too.
Implementing the same MiniTwit web-application in various programming languages, with different web-frameworks, and on different versions of compilers/interpreters.

Thank you!

On the Energy Consumption of CPython
Helge Pfeiffer
ropf@itu.dk
Research Centre for Government IT
IT University of Copenhagen