The batch compose and read benchmarks in this section are written using
``perf`` library, created by Viktor Stinner. For more information on how to get
reliable results of test runs please consult
http://pyperf.readthedocs.io/en/latest/run_benchmark.html.

The `simple_` benchmarks can be just run, consult command line argument on how
to run those.


For profiling speed degradation use either::

    python -m cProfile -o profile.dump benchmark/simple_produce_bench.py
    python -m snakeviz profile.dump

Or if the problem is not visible, because most of the time we spent in epoll(),
you can use some static profiler like vmprof. Install the server or use
vmprof.com. For local server setup see https://github.com/vmprof/vmprof-server

To run using vmprof.com for visualization::

    python -m vmprof --web benchmark/simple_produce_bench.py

For local server::

    python -m vmprof --web-url http://localhost:8000 --web benchmark/simple_produce_bench.py