Announcing the release of pprocess 0.3 (previously known as parallel/ pprocess), available from...
Note that only POSIX-like platforms are supported in this release, although improvements in portability and all other areas are welcome.
What is it? -----------
The pprocess module provides several mechanisms for running Python code concurrently in several processes. On systems with multiple CPUs or multicore CPUs, processes should take advantage of as many CPUs or cores as the operating system permits.
Several different abstractions are supported by the module, including channels (and communicating sequential processes), queues and a parallel map function. The module gives developers a certain amount of flexibility in developing new abstractions, and a motivation behind the development of the module was, in fact, to discover which abstractions are most convenient when writing parallel-aware programs.
In order to get a feel for the abstractions and mechanisms provided by the module, a tutorial has been made available; it illustrates the process of converting certain kinds of sequential programs into parallel programs:
Future Work -----------
The module has been tested with non-trivial programs (see the PyGmy raytracer example in the distribution), but performance results have been varied: pprocess ensures the utilisation of many processors but could arguably show better scalability. It is envisaged that the focus for future releases involve improvements to the interprocess communications mechanisms employed, along with an investigation into more effective ways of distributing computations in specific programs.
In the context of recent discussions about Python concurrency support , this module is intended as a demonstrator for various API conveniences, with an aim of encouraging standardisation and potentially the eventual standard library adoption of a useful parallel processing module.