Next: Parallel building blocks
Up: R. Clapp: Python and
Previous: Options, Flows, and Programs
The library is currently designed for coarse-grain parallel
jobs that fit on a single
processor (no inter-process communication necessary).
The user writes a serial code that works on a portion or all
of the dataset.
Each parallel job is broken up into a series of tasks.
These tasks can have the same, or different parameters, and
the various input and output files can be distributed in
several different manners.
There are three basic classes of parallel objects: core objects
that handle communication, parallel file objects that
describe how a given file is distributed and parallel job
objects that handle distributing the various tasks.