energy_module

Energy Modules

This directory contains scripts, wrappers, and configuration files used to measure CPU energy consumption during the execution of Python benchmarks in our study: "Python Under the Microscope: A Comparative Energy Analysis of Execution Methods."

All energy readings are captured via Intel RAPL counters using the pyRAPL library, providing precise, reproducible, and software-level energy profiling on Intel-based systems.

Directory Structure

energy_modules/
├── __pyRAPL__/
│   ├── measure_pyrapl.py       # Benchmark wrapper with pyRAPL decorator
measurement
└── README.md                   # This file

Module Description

Module	Description
measure_pyrapl.py	Core script used to wrap and execute a benchmark under energy measurement.
pyRAPL	Python library for accessing Intel RAPL (Running Average Power Limit) counters.
setup_pyrapl.md	Contains step-by-step setup for configuring permissions and dependencies.

Features

• Uses Intel RAPL to read energy usage in microjoules (μJ) for the CPU package.
• Integrates with your benchmark using Python decorators, ensuring low overhead.
• Logs energy measurements in CSV and JSON for easy downstream analysis.
• Supports batch execution and repeat measurement with consistent sampling conditions.
• Focuses on runtime-only power usage (excludes build/compilation energy).

⚠️ Only works on Intel-based Linux machines with RAPL access.

Setup Instructions

Follow the setup guide in setup_pyrapl.md to:

1. Install pyRAPL:

   pip install pyRAPL

2. Ensure access to /sys/class/powercap/ (requires root or permissions fix):

   sudo chmod -R a+r /sys/class/powercap/intel-rapl:*/energy_uj

3. (Optional) Set CPU affinity and disable turbo boost for consistent readings.

Example Usage

To measure the energy usage of a benchmark (e.g., binary_tree.py):

cd energy_modules/__pyRAPL__/

nice -n -20 ionice -c2 -n0 python measure_pyrapl.py ../../benchmarks/Binary-Trees/CPython/binary_tree.py

This command:

• Gives high CPU and I/O priority to reduce external noise.
• Executes the benchmark wrapped in a pyRAPL measurement context.
• Saves energy logs to a structured CSV file for post-analysis.

Output Format

• Output is saved as both .csv and .json.

• Fields include:

  • timestamp
  • benchmark name
  • energy (μJ)
  • duration (s)
  • package/core/domain (if available)

Repetition and Batch Execution

In experiments, each benchmark was run 50 times to account for natural fluctuations and background processes. You can modify the wrapper to include repetitions or batch folder traversal.

Notes on Accuracy

• RAPL provides hardware-level counters for energy estimation.
• We measure only CPU Package Power as DRAM counters are often unavailable on laptops.
• Results were averaged across 50 runs and normalized per algorithm for fair cross-method comparison.