docker speeds up python, why?

Question

So I am playing with docker for the first time on my mac. I used boot2docker through the standard tutorial and I am starting a prompt in an ubuntu image.

docker pull ubuntu
docker run -i -t ubuntu /bin/bash

When in docker I started my first experiment, to see if the performance would go down. From the command line I would use the python timeit module to quickly check some basic performance measures.

Mac Python Results

$ python3.4 -m timeit '"-".join(str(n) for n in range(100))'
10000 loops, best of 3: 37.7 usec per loop
$ python3.4 -m timeit '"-".join([str(n) for n in range(100)])'
10000 loops, best of 3: 34.2 usec per loop
$ python3.4 -m timeit '"-".join(map(str, range(100)))'
10000 loops, best of 3: 26.2 usec per loop

Docker Python Results

> python3 -m timeit '"-".join(str(n) for n in range(100))'
10000 loops, best of 3: 30 usec per loop
> python3 -m timeit '"-".join([str(n) for n in range(100)])'
10000 loops, best of 3: 26.9 usec per loop
> python3 -m timeit '"-".join(map(str, range(100)))'
10000 loops, best of 3: 20.2 usec per loop

It seems strange that the docker ubuntu, that is running on top of my mac, is actually running python code faster than the python on mac. Is there any reason for why this might be?

Edits

I can confirm that both python versions are running in 64 bit.

Mac Python

python3 -c 'import sys;print("%x" % sys.maxsize, sys.maxsize > 2**32)'
7fffffffffffffff True

Ubuntu Python

python3.4 -c 'import sys;print("%x" % sys.maxsize, sys.maxsize > 2**32)'
7fffffffffffffff True

Answer 1

This is more about the difference in operating systems than about docker performance. Measuring performance of applications can be tricky.

Bottom line is that OS X has a good number of processes that will compete with your test and OS X likely did not give your test a high priority.

A container should perform as well as the native environment (sometimes better) in most cases. But, your test should make the container do work. Docker will need to add overhead when your application is making system calls and accessing I/O, so those should both be included in your test.

IBM wrote a paper on the Linux Container vs. Native env issue last year.

http://domino.research.ibm.com/library/cyberdig.nsf/papers/0929052195DD819C85257D2300681E7B/$File/rc25482.pdf