Epilogue: Global Interpreter Lock (GIL)

The standard Python interpreter (called CPython) does not support the use of threads well. This is because it, and by association all C/C++/Fortran-based modules/extensions to CPython have all been written to assume that an individual Python script is serial (i.e. it only has a single thread of execution). The CPython Python interpreter uses a “Global Interpreter Lock” to ensure that only a single line of a Python script can be interpreted at a time, thereby preventing memory corruption caused by multiple threads trying to read, write or delete memory in parallel. This means that, even if you use the Python threading module, you will still only execute a single line of your script at a time in CPython. There are attempts and discussions aimed at removing the GIL from CPython, but it is a task that is extremely difficult to achieve.

Because of the GIL, parallel Python is normally based on running multiple forks of the Python interpreter, each with their own copy of the script and their own GIL. Every time a script needs to run in parallel, the Python interpreter is forked into multiple processes, with each forked process performing their part of the shared work. Once the parallel work is complete (e.g. the multiprocessing.Pool is terminated), the forked processes are killed.

The multiprocessing module solves the problem of the GIL, but at the cost of very high overhead (multi-milliseconds) for entering and leaving each parallel section of code, and the higher cost of sharing data between workers as compared to a true multi-threaded program.

However, this higher overhead is not a true problem for most scripts. If the overhead it too high, and greater performance is desired, then the “slow” parts of the script can be rewritten in a compiled language such as C, C++ or Fortran, and linked in to Python as a CPython extension. If performance is still a problem, then these CPython extensions can then be parallelised themselves using OpenMP, MPI etc. (there is nothing stopping a CPython extension from running its own code in parallel, as long as it doesn’t try to back-call into the CPython interpreter).

If you want more information on how to write CPython extensions in C++, check out this tutorial.


Previous Up Next