Epoll examples and benchmarks
Simple epoll based HTTP servers with constant JSON response to any request.
It's possible to make and run examples on your system as well as in Docker.
To run C version in the Docker container use the next commands:
# cd c/
# docker build -f Dockerfile_C -t epoll .
# docker run --net=host -p 8000:8000 -t epoll
After that you'll see the next output:
Successfully tagged epoll:latest
Server listening on 0.0.0.0:8000
For benchmarks wrk tool was used.
NB: Even with --net=host key in docker run command the performance of examples on a host system is better than in the Docker.
To test I've used the next wrk confuguration: 2 threads, 100 clients and 10 seconds.
Tested epoll servers were run in Docker, whereas wrk was on the host system (Ubuntu 16.04).
C (threaded_loop (2 threads)):
$ wrk -t2 -c100 -d10s http://0.0.0.0:8000/
Requests/sec: 77176.69
=============================================
CPython3 (multiprocessed_loop (2 processes)):
$ wrk -t2 -c100 -d10s http://0.0.0.0:8000/
Requests/sec: 48215.67
CPython3 (threaded_loop (2 threads)):
$ wrk -t2 -c100 -d10s http://0.0.0.0:8000/
Requests/sec: 29849.85
=============================================
PyPy3 (multiprocessed_loop (2 processes)):
$ wrk -t2 -c100 -d10s http://0.0.0.0:8000/
Requests/sec: 76399.70
PyPy3 (threaded_loop (2 threads)):
$ wrk -t2 -c100 -d10s http://0.0.0.0:8000/
Requests/sec: 74526.86
=============================================
Go (ConcurrentLoop (2 threads)):
$ wrk -t2 -c100 -d10s http://0.0.0.0:8000/
Requests/sec: 76362.73
It's interesting that PyPy3 performance is close to C.
Released under The MIT License.