Programming for High Performance
                        An Introduction

Many interesting problems in science and technology require massive
computational power to investigate them. Gene analysis, weather
prediction, and wind tunnel simulations are popular examples. High
performance computing facilities are located in computing centers of
universities and research laboratories and are not as easily
accessible and usable as workstations. The concepts of how to use them
efficiently are therefore not common knowledge.

This lecture will give an introduction into programming for high
performance. We will at first look at the hardware aspects of modern
supercomputers and classify their architectural
characteristics. Second, paradigms of parallelization will be
discussed. The concrete way to bring a program to a parallel computer
depends on the architecture of the machine, the structure of the
problem, and the programming concept chosen. Usually, the first two
factors are fixed and we are free to select our favorite programming
concept in dependence of both of them. Therefore, we will discuss
programming concepts in detail. We can distinguish message passing
based programming, shared memory based programming and the
data-parallel programming model. The lecture will give an
introduction into all three of them and compare their applicability.

At the end of the lecture the student should be able to answer the
following question: I have a given problem and my computing center has
a high performance computer. How do I write a program to use this
resource efficiently?

The lecture is oriented towards computer scientists and researchers
in applied sciences who want to introduce high performance computing
to their field. The presented material is also a good starting point
for those who are interested in architectural aspects of parallel
hardware and in aspects of parallel programming languages.

Requirements: Basic programming capabilities.

Slides will be available for printing.