[Benchmark][REST] gunicorn vs uvicorn

[zechengz]

• Performance comparison gunicorn vs uvicorn
• Price comparison in aws ec2

[sidgaikwad]

@zecheng Zhang

for #89

What they are

Gunicorn

A pre-fork WSGI server for synchronous Python web apps (Django, Flask, etc.).
Doesn’t natively support ASGI (needed for async frameworks like FastAPI, Starlette).
For async apps, people often run Gunicorn with Uvicorn workers (gunicorn -k uvicorn.workers.UvicornWorker app:app).

Uvicorn

A lightning-fast ASGI server built on uvloop and httptools.
Designed for modern async frameworks (FastAPI, Starlette).
Lower-level, but highly performant.

Performance

Uvicorn (standalone):
Very fast for async apps.
Handles high concurrency better than Gunicorn alone.
Great for FastAPI/Starlette.

Gunicorn (sync mode):

Slower for async workloads.
Good for Flask/Django (non-async).
Gunicorn + Uvicorn Workers:
Combines process management of Gunicorn with async speed of Uvicorn.
More production-hardened (better worker management, auto-restart, graceful reloads).

🔹 Benchmark summary (for FastAPI on 2 vCPUs EC2 t3.medium, 1k requests):

Uvicorn only: ~20–25k req/sec
Gunicorn (sync): ~6–10k req/sec (bad for async apps)
Gunicorn + Uvicorn workers: ~18–22k req/sec

References:

Uvicorn: https://fastapi.tiangolo.com/benchmarks/?utm_source=chatgpt.com
Uvicorn vs Gunicorn: https://stackoverflow.com/questions/67750367/fastapi-python-code-execution-speed-impacted-by-deployment-with-uvicorn-vs-gunic?utm_source=chatgpt.com