A Really Simple Benchmark Of Some Web Frameworks

• Basic servers for Python, NodeJS, and Rust, each implemented to run in docker
  • Python: FastAPI + uvicorn
  • NodeJS: ExpressJS
  • Rust: Rocket.rs
• Each server is set up with a single route. When called, the server counts from 0 to 100,000, then returns a json response.
• Each server is set up to listen on host port 8000
  • Each folder contains a docker-compose.yml
  • Run docker-compose up --build to start the server
  • Run docker-compose down to stop it
• We have a shell script ./curl_loop.sh on the docker host
  • This script generates a curl command to call a server 1000 times in a single http connection (attempting to reduce both bash and http overhead)
  • It then prints out the execution time of the script
• Each time we start a different server for testing, we only keep the timing for the 2nd run of the curl_loop.sh. In theory, this should exclude the time for whatever jit compilation that the frameworks might perform.

Results

Tests are run on a single linux laptop, on this particular config of hardware and software that we shall not concern ourselves with.

• Python: ~2 seconds

  real	0m2.014s
  user	0m0.045s
  sys	0m0.026s
  

• NodeJS: ~0.23 seconds

  real	0m0.233s
  user	0m0.028s
  sys	0m0.025s
  

• Rust: ~0.09 seconds

  real	0m0.091s
  user	0m0.035s
  sys	0m0.016s