Faster Is Better

Python apps on lower-end hardware like Raspberry Pis can be a bit slow, but luckily you have some options that can improve their performance. A number of interesting packages allow you to compile, interpret, or repackage your Python apps. In this article, I highlight two packages that have given me good success:

• PyPy [1] – a replacement to the native Python interpreter that runs more than four times faster
• Nuitka [2] – a native Python utility to compile Python apps to C code

How a Python application performs is based on a lot of different factors, so it's important to know the limitations and how best to work with them. PyPy and Nuitka might not work for all applications, but for many common types of projects they are great fits.

Background

PyPy has been around for a while, and Guido van Rossum, the creator of Python, is popularly said to have stated: "If you want your code to run faster, you should probably just use PyPy."

The PyPy site has some performance results, and they state that, on average, PyPy will run 4.2 times faster than native Python. In one of my test projects, it ran nine times faster than native Python. PyPy really shines in the area of web services, database connections, and iterative (large loop) applications.

It's very important to note that PyPy only supports a limited number of Python libraries. For Raspberry Pi projects, two of the important libraries that PyPy does not support are tkinter, a graphic interface library, and RPi.GPIO, the Raspberry Pi general purpose I/O library. To see whether PyPy will work with your application, check the supported libraries [3].

Nuitka will not give the same performance results as PyPy, but it can offer some speed improvements over native Python. Nuitka supports a large majority of the Python libraries, so it's a good fit when you can't use PyPy. Nuitka has the added benefit of creating a compiled application, so you don't need to distribute Python source code.

Figure 1 summarizes the project requirements best served by PyPy and Nuitka.

Figure 1: Best fits for PyPy and Nuitka.

PyPy – A Faster Python

The improved performance of PyPy is due to its just-in-time compiler, as opposed to the native Python's line-by-line interpreter.

PyPy has a version for Python 2.7 and 3.6 for Linux, macOS, and Windows. The PyPy version for Python 2.7 is preinstalled on the latest Raspbian operating system for the Raspberry Pi. To install the PyPy3 version in Ubuntu, enter:

sudo apt update
sudo apt install PyPy3

The nice thing about PyPy is that you can use your base Python code as is, and you can do basic testing with PyPy in command-line mode. For example, on the Raspberry Pi, if I want to check whether some basic Python libraries are loaded and then see if the Pi GPIO library is supported, I would use the pypy3 command (Listing 1).

§§noumber
 $ pypy3
Python 3.5.3 (7.0.0+dfsg-3, Mar 03 2019, 06:11:22)
[PyPy 7.0.0 with GCC 8.2.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>>> import os
>>>> import time
>>>> import RPi.GPIO
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/usr/lib/python3/dist-packages/RPi/GPIO/__init__.py", line 23, in <module>
    from RPi._GPIO import *
ImportError: No module named 'RPi._GPIO'
>>>>

To run your Python application from the command line, substitute python with PyPy (or PyPy3):

$ PyPy3 myapp.py

If you are running your Python app as an executable script (i.e., as chmod +x myapp.py), then you'll need to change the first line of your script from #!/usr/bin/python to #!/usr/bin/PyPy3.

PyPy Libraries

By default, only the basic Python libraries are installed with PyPy. To load a specific Python library into PyPy3, you need to install the Python package installer (pip) module before Python packages can be loaded:

$ wget https://bootstrap.pypa.io/get-pip.py
$ PyPy3 get-pip.py
$ PyPy3 -m pip install <some-pymodule>

I found I was able to load some of the "lighter" modules (e.g., bottle, requests, beautifulsoup4, pika, etc.) without any issues. However, some of the "heavier" modules (e.g., NumPy and MatPlotlib) would not load directly. If you're planning on using some of these modules, the recommendation is to run PyPy in an isolated Python environment (virtualenv) [4].