260008 VO+UE Parallel programming (2008S)

Although "parallel computing" has been around as a buzzword since the late 1970s, practical acceptance of the concept is lagging far behind the original expectations. In the past this may have been partly due to often prohibitive hardware costs, but the primary reasons probably are the non-emergence of a coherent line of development and the concomitant lack of genuine software standards.
Potential users reluctant to commit themselves to a specific architecture bound to become a dead end in a few years time, may resort to traditional high-level programming languages augmented by parallel constructs, such as HPF (High Performance Fortran), Fortran 90/95, OpenMP, etc. Of course, this is tied to the availability of a "parallelizing compiler" adapted to the hardware in question. On the other hand, in recent years, two packages of communication libraries (PVM - Parallel Virtual Machine and MPI - Message Passing Interface) have gained wide-spread acceptance. Using calls to these library procedures, sufficiently "coarse-grained" applications may be readily parallelized on low-cost workstation or PC (Beowulf) clusters.
Illustrated by a series of simple examples, this course provides a compact overview of MPI and OpenMP and introduces a number of basic parallelizing strategies. Participants are given the opportunity to implement mini-projects of their choice on a PC cluster.

Prerequisites: good working knowledge of Fortran and/or C; some Unix.