AsyncWebSocket is ~4.5x slower than AIOHTTP #645
Description
Overview
The asyncio websocket implementation is approximately 4.5x slower compared to AIOHTTP.
Curl-CFFI:
AIOHTTP:
Results
I am running a fresh install of Ubuntu 24.04.3 LTS (Noble), running bare metal on an HPE DL-60 Gen9 server.
aiohttp 3.12.15
├── aiohappyeyeballs >=2.5.0
├── aiosignal >=1.4.0
│ └── frozenlist >=1.1.0
├── attrs >=17.3.0
├── frozenlist >=1.1.1
├── multidict >=4.5,<7.0
├── propcache >=0.2.0
└── yarl >=1.17.0,<2.0
├── idna >=2.0
├── multidict >=4.0
└── propcache >=0.2.1
curl-cffi 0.13.0
├── certifi >=2024.2.2
└── cffi >=1.12.0
└── pycparser *
uvloop 0.21.0
Results on my machine:
--- Starting curl-cffi Benchmark ---
Connection established (curl-cffi). Starting data transmission...
--- curl-cffi Benchmark Complete ---
Sent 2.00 GB of data in 119.61 seconds.
Average throughput: 0.13 Gbps.
Counter (curl-cffi) returned: 0.42 GB
========================================
--- Starting aiohttp Benchmark ---
Connection established (aiohttp). Starting data transmission...
--- aiohttp Benchmark Complete ---
Sent 2.00 GB of data in 26.48 seconds.
Average throughput: 0.60 Gbps.
Counter (aiohttp) returned: 2.05 GB
I wrote a basic benchmark which you can use to test this. It's a simple server which sends a continnous stream of data back and forth. It's not good code but it illustrates the point. I am running this benchmark with the server and client both using localhost so that real network conditions are not a factor.
The server/client is using TLS since that's a representative use case. On my machine there is not much difference between the TLS version and non-TLS version, so I am only including the TLS version.
Dependencies
I used a Poetry venv so that other system packages don't cause issues.
In order to reproduce the same environment, you can use this pyproject.toml file:
[project]
name = "ws-bench"
version = "0.1.0"
description = ""
authors = [
{name = "Your Name",email = "[email protected]"}
]
license = {text = "MIT"}
readme = "README.md"
requires-python = ">=3.13"
dependencies = [
"curl-cffi (>=0.13.0,<0.14.0)",
"aiohttp (>=3.12.15,<4.0.0)",
"uvloop (>=0.21.0,<0.22.0)"
]
[tool.poetry]
package-mode = false
[build-system]
requires = ["poetry-core>=2.0.0,<3.0.0"]
build-backend = "poetry.core.masonry.api"Install it (same directory) with:
sudo apt install pipx
pipx install poetry
pipx ensurepath
exec "$SHELL"
poetry install
eval $(poetry env activate)Server
In order to run this, you need to start by generating a cert/key:
openssl req -x509 -newkey rsa:2048 -nodes -keyout localhost.key -out localhost.crt -days 365 -subj "/CN=localhost"Here is the server code: server_ssl.py
Run with:
chmod +x server_ssl.py
./server_ssl.pyClient
Here is the client code: client.py
Run with:
chmod +x client.py
./client.py