High-level parallelization in Python
Konrad Hinsen
hinsen@cnrs-orleans.fr
Tue, 30 Oct 2001 19:04:14 +0100
The current development version of Scientific Python contains an
experimental package for high-level parallelization. It is based on
the Bulk Synchronous Parallel (BSP) paradigm (see
www.bsp-worldwide.org for more information), which is much simpler
than the more popular message-passing paradigm and also safer,
especially due to the impossibility of deadlocks.
The main initial goal of the Python BSP package is a well-integrated
Pythonesque implementation of BSP that makes parallel programming
accessible to a wider range of developers. Performance has been a
secondary criterion until now and can certainly be improved, but the
current implementation is quite usable for coarse-grained parallelism.
The package is labelled "experimental" because it has only been tested
with a very small number of applications; future experience may lead
to modifications that are not necessarily fully compatible with the
current release.
Scientific Python 2.3.1 is available from
ftp://dirac.cnrs-orleans.fr/pub/ScientificPython/ScientificPython-2.3.1.tar.gz
Prerequisites are Python 1.5.2 or higher (2.1 recommended for full
access to all features) and an MPI implementation. A tutorial is
included, but can also be obtained separately from
http://dirac.cnrs-orleans.fr/programs/BSP_Tutorial.pdf
--
-------------------------------------------------------------------------------
Konrad Hinsen | E-Mail: hinsen@cnrs-orleans.fr
Centre de Biophysique Moleculaire (CNRS) | Tel.: +33-2.38.25.56.24
Rue Charles Sadron | Fax: +33-2.38.63.15.17
45071 Orleans Cedex 2 | Deutsch/Esperanto/English/
France | Nederlands/Francais
-------------------------------------------------------------------------------