The goal of the presentation is to give an overview about how to build a
commodity PC based GNU/Linux cluster for High Performance Computing
(HPC) in research. After the presentation, the audience should be able
to decide whether a source based GNU/Linux distribution is suitable for
their computing needs, and how to actually install a source based
computing cluster on their own.

Due to the extreme flexibility of the GNU/Linux operating system and the
large variety of hardware components, building a cluster for High
Performance Computing (HPC) is still a challenge in many cases. At the
Division of Intelligent Systems at the Norwegian University of Science
and Technology (NTNU), we have build a 40 node HPC cluster for research
purposes using the source-based GNU/Linux distribution SourceMage.

We describe a methodology for designing and installing a highly
customized GNU/Linux cluster. Different types of Linux distributions
will be mentioned, binary-based and source-based, with their advantages
and disadvantages.

The presentation will focus on using SourceMage for HPC, specifying the
'magical' ideas behind it: the ease of upgrading to the latest available
version of the source code, a packaging system for keeping track of
dependencies, optimized compiles for the hardware architecture used,
easy integration of new packages, amongst others.

*Topics*

* Why can a Beowulf-type PC cluster be used for High Performance
Computing needs and how does that differ from other types of distributed
and parallel computing environments?

* Which type of hardware and software can be used?

* Software necessary to run a PC cluster, usage of standard services.

* Which libraries and tools can be used for parallel and distributed
computing?

* User management and security issues, job scheduling and load
distribution technologies.

* Aspects of node configuration and different methods for node cloning.

* Cluster installation and administration.

* Different concepts of booting, running, and maintaining the nodes.

* The special advantages and pitfalls of choosing a source based
software installation and maintenance compared to a binary based solution.

* Benefits and problems in customizing the Linux kernel and other
aspects of the operating system environment.


*Intended Audience*

Intended for scientists and engineers interested in learning about ways
to set up a HPC cluster. Neither in depth knowledge of the Linux
operating system, nor about clustering technologies is necessary.
General knowledge about Unix type systems and parallel or distributed
computing is helpful.