The pyina package provides several basic tools to make MPI-based high-performance computing more accessible to the end user. The goal of pyina is to allow the user to extend their own code to MPI-based high-performance computing with minimal refactoring.
The central element of pyina is the parallel map-reduce algorithm. Pyina currently provides two strategies for executing the parallel-map, where a strategy is the algorithm for distributing the work list of jobs across the available nodes. These strategies can be used “in-the-raw” (i.e. directly) to provide map-reduce to a user’s own mpi-aware code. Further, pyina provides the “ez_map” interface, which is a map-reduce implementation that hides the MPI internals from the user. With ez_map, the user can launch their code in parallel batch mode – using standard python and without ever having to write a line of parallel python or MPI code.
There are several ways that a user would typically launch their code in parallel – directly with “mpirun” or “mpiexec”, or through the use of a scheduler such as torque or slurm. Pyina encapsulates several of these ‘launchers’, and provides a common interface to the different methods of launching a MPI job.
Pyina is part of pathos, a python framework for heterogenous computing. Pyina is in the early development stages, and any user feedback is highly appreciated. Contact Mike McKerns [mmckerns at caltech dot edu] with comments, suggestions, and any bugs you may find. A list of known issues is maintained at http://trac.mystic.cacr.caltech.edu/project/pathos/query.
Pyina provides a highly configurable parallel map-reduce interface to running MPI jobs, with:
- a map-reduce interface that extends the python 'map' standard - the ability to submit batch jobs to a selection of schedulers - the ability to customize node and process launch configurations - the ability to launch parallel MPI jobs with standard python - ease in selecting different strategies for processing a work list
The latest stable release version is pyina-0.1a1. You can download it here. The latest stable version of pyina is always available at:
If you like living on the edge, and don’t mind the promise of a little instability, you can get the latest development release with all the shiny new features at:
or even better, fork us on our github mirror of the svn trunk:
Pyina is packaged to install from source, so you must download the tarball, unzip, and run the installer:
[download] $ tar -xvzf pyina-0.2a.dev.tgz $ cd pyina-0.2a.dev $ python setup py build $ python setup py install
You will be warned of any missing dependencies and/or settings after you run the “build” step above. Pyina depends on dill and mpi4py, so you should install them first. A version of MPI must also be installed. Pyina’s launchers that submit to a scheduler will throw errors if the underlying scheduler is not available, however a scheduler is not required for pyina to execute.
The current implementation of pyina required slight modifications to one external dependency. This dependency and installation instructions are provided in the pyina.external directory for users installing without setuptools.
Alternately, pyina can be installed with easy_install:
[download] $ easy_install -f . pyina
- python, version >= 2.5, version < 3.0 - mpi4py, version >= 1.2.1 - dill, version >= 0.2a.dev
- setuptools, version >= 0.6 - pyre, version == 0.8 - mystic, version >= 0.2a2.dev
Probably the best way to get started is to look at a few of the examples provided within pyina. See pyina.examples for a set of scripts that demonstrate the configuration and launching of mpi-based parallel jobs using the ez_map interface. Also see pyina.examples_other for a set of scripts that test the more raw internals of pyina.
Important classes and functions are found here:
- pyina.pyina.ez_map [the map-reduce API definition] - pyina.pyina.mappers [all available strategies] - pyina.pyina.schedulers [all available schedulers] - pyina.pyina.launchers [all available launchers]
Mapping strategies are found here:
- pyina.pyina.parallel_map [the card-dealer strategy] - pyina.pyina.parallel_map2 [the equal-portion strategy]
Pyina also provides two convience scripts that help navigate the MPI environment. These scripts are installed to a directory on the user’s $PATH, and thus can be run from anywhere:
- machines.py [list the available MPI nodes] - mpi_world.py [setup/teardown of the MPI environment]
If may also be convienent to set a shell alias for the launch of ‘raw’ mpi-python jobs. Set something like the following (for bash):
$ alias mpython1='mpiexec -np 1 `which python`' $ alias mpython2='mpiexec -np 2 `which python`' $ ...
Pyina is distributed under a 3-clause BSD license.
>>> import pyina >>> print pyina.license()
If you use pyina to do research that leads to publication, we ask that you acknowledge use of pyina by citing the following in your publication:
M.M. McKerns, L. Strand, T. Sullivan, A. Fang, M.A.G. Aivazis, "Building a framework for predictive science", Proceedings of the 10th Python in Science Conference, 2011; http://arxiv.org/pdf/1202.1056 Michael McKerns and Michael Aivazis, "pathos: a framework for heterogeneous computing", 2010- ; http://trac.mystic.cacr.caltech.edu/project/pathos
Please see http://trac.mystic.cacr.caltech.edu/project/pathos or http://arxiv.org/pdf/1202.1056 for further information.
TODO: Figure out how to actually get changelog content.
Changelog content for this version goes here.