aiohttp.TCPConnector(limit=16)- alpine linux
- nginx 1.21
- python 3.9
- aiohttp dev-master
- chardet 4.0.0 (aiohttp-chardet)
- charset-normalizer 2.0.6 (aiohttp-next)
Prepare the environment
docker-compose build aiohttp-next server
docker-compose build --no-cache aiohttp-chardet
docker-compose up -d server
Collect results
docker-compose up aiohttp-chardet
docker-compose up aiohttp-next
Memory sampling (to be run in parallel of aiohttp-* services)
docker stats
This come with no surprise as it was already proven previously.
- using charset-normalizer
- 37.7s with 467 files --> 81ms
- using chardet
- 189.3s with 467 files --> 405ms
5 times faster while keeping 88% (410 / 464 files) full backward compatible results.
Following issue aio-libs/aiohttp#4112 it could be a good idea to showcase that aspect.
Using docker stats for the sampling.
Bellow, the peak usage for each flavor.
- using charset-normalizer
- 38MiB
- using chardet
- 67MiB
1.7 times less memory consumption. For a 5 times faster guess output (in avg).
Keep in mind that the actual delays may vary depending on your CPU capabilities. The factor between chardet/charset-normalizer flavors should remain intact.