EcoTrace Academic Benchmark Suite

Scientific, reproducible green computing benchmarks evaluating energy efficiency, operational carbon emissions ($gCO_2eq$), and hardware power scaling across diverse computing workloads.

---

Benchmark Index & Studies

Study	Title	Category	Focus	Runner
01	Pandas vs. Polars	Data Engineering	Vectorization, Arrow format, Race-to-Sleep	python pandas_vs_polars.py
02	PyTorch vs. TensorFlow	Machine Learning	CNN Training on CIFAR-10, Carbon per Accuracy Point	python 02_ml_training_frameworks.py
03	Sorting Algorithm Complexity	Computer Science	$O(n \log n)$ vs $O(n^2)$ physical energy scaling	python 03_sorting_algorithms.py
04	Web Framework Efficiency	Cloud Systems	Flask vs FastAPI request carbon cost under load	python 04_web_frameworks.py
05	LLM Inference Scaling	Generative AI	Per-token energy cost across GPT-2 parameter tiers	python 05_llm_inference.py
06	Cross-Region Grid Variability	Cloud Architecture	Up to 70x spatial carbon difference across 15 countries	python 06_regional_carbon.py
07	Accuracy vs. RAPL	Energy Metrology	Model-vs-hardware validation (MAPE, MAE, R²) — Linux/RAPL for ground truth, synthetic fallback elsewhere	python validation/accuracy_vs_rapl.py

---

Framework Architecture (benchmarks/framework/)

All studies in this suite leverage our reusable benchmarking framework: - BenchmarkRunner: Standardized harness managing warm-up iterations, cooldown intervals, and automated EcoTrace telemetry collection. - BenchmarkStatistics: Statistical aggregator providing 95% confidence intervals, IQR outlier rejection, and Mann-Whitney U hypothesis significance tests ($p < 0.05$). - EnvironmentSnapshot: Captures hardware fingerprints (CPU, GPU, RAM, power state) and package dependency trees to ensure scientific reproducibility.

---

Quickstart

# 1. Install benchmark dependencies
pip install -r benchmarks/requirements.txt

# 2. Execute any benchmark
python benchmarks/03_sorting_algorithms.py

# 3. Results are saved to benchmarks/results/<study>.json

# 4. (Optional) Regenerate the article from the latest results
python benchmarks/report_generator.py 03
# Or regenerate all registered studies at once:
python benchmarks/report_generator.py

The report generator reads benchmarks/results/*.json and rewrites the corresponding article in benchmarks/articles/*.md. Run it after every benchmark to keep the published numbers in sync with the actual measurements.

---

Citation

If you use EcoTrace or these benchmarks in academic research, please cite:

@software{ozkal2026ecotrace,
  author = {Ozkal, Emre},
  title = {EcoTrace: High-Precision Energy and Emissions Instrumentation for Python},
  year = {2026},
  url = {https://github.com/Zwony/ecotrace},
  version = {1.6.0}
}