Skip to content

Carbon Cost of Serving: Web Framework Throughput vs. Energy per Request

Comparing fastapi, flask under identical synthetic HTTP load.

Auto-generated from benchmarks/results/*.json on 2026-09-05 13:17:32 UTC. Re-run python benchmarks/report_generator.py to refresh after new measurements.

Executive Summary

Different Python web frameworks have measurably different energy profiles per request under identical load. This study benchmarks the carbon cost of serving a simple JSON endpoint under a synthetic concurrent workload.

1. Methodology

  • Workload: Simple JSON API endpoint returning a fixed payload (20 items)
  • Concurrent requests per run: 10
  • Total requests per run: 5000
  • Measured runs per framework: 3
  • Load generator: Python urllib with ThreadPoolExecutor

2. Results

Framework Mean Duration (s) Mean Carbon (gCO2) Req/s (mean) p50 Latency (ms) p99 Latency (ms)
FLASK 18.9725 s [15.3755 s, 22.5695 s] 74.8 mg [59.7 mg, 89.9 mg] 265 35.2 97.2
FASTAPI 22.5036 s [17.8062 s, 27.2009 s] 81.5 mg [59.9 mg, 103.1 mg] 199 46.0 162.0

3. Pairwise Comparison

  • Baseline: flask
  • Challenger: fastapi
  • Speedup ratio: 0.84x
  • Carbon reduction: -8.9%
  • Statistical significance (duration): NO (p = 0.1266)

4. Reproducibility

cd benchmarks
python 04_web_frameworks.py
python report_generator.py 04

Output JSON: benchmarks/results/04_web_frameworks.json

Test Environment

  • CPU: 13th Gen Intel(R) Core(TM) i7-13700H
  • Clock: 2.9180 GHz
  • Cores: 14 physical / 20 logical
  • Memory: 15.65 GB
  • OS: Windows 10
  • Python: 3.11.1
  • Power source: AC
  • Run timestamp: 2026-09-05T13:14:07.699185+00:00
  • Key packages:
  • ecotrace: 1.5.1
  • fastapi: 0.111.0
  • flask: 3.1.0
  • matplotlib: 3.10.9
  • numpy: 2.4.6
  • pandas: 3.0.5
  • polars: 1.44.1
  • psutil: 7.2.2
  • torch: 2.13.0+cpu
  • uvicorn: 0.30.1