Fastest European made supercomputer inaugurated

the University of Paderborn is the first and largest customer of the hpcLine. Fujitsu Siemens has, however, sold several smaller models to other universities, and expects to conclude a sale to an industrial research department in the near future.

Christian Bischof, director of the computing centre of the University of Technology in Aachen, showed several examples of technical-scientific applications of the hpcLine at his university. Main application area there is fluid dynamics. Currently the cluster runs Redhat Linux, but they are also looking at Suse Linux.

In Aachen, they are interested in running applications on the hpcLine and not, in contrast to Paderborn, in computer science and parallel processing as research topics. That is why Bischof would like to see more commercial packages ported to the hpcLine. However, the GNU and PGI compilers on the machine perform very well and provide a good basis to port or develop applications.

According to Bischof, a Linux cluster requires still a lot of system administration work. Also the message passing paradigm one needs to understand to be able to program the machine, is more difficult to understand and handle than the vector paradigm or the shared memory paradigm.

The hpcLine in Paderborn is mainly used to develop new parallel applications and to do research into parallel processing. Two examples are air line scheduling and remote rendering.

In airline crew scheduling, the problem to be solved for instance, is to get the right planes and the right crew at the right time on the right airport to start a flight. This is a very complex task for a large company when also sleeping time and holidays for the crew have to be taken into account. Especially when there are delays, for instance due to technical problems, or because of delays at an airport, a fast answer to a possible redistribution and rescheduling is needed.

Rendering of images, calculating the 2-D images to present on a screen from a 3-D model, is a very compute intensive problem. One of the techniques used is the "radiosity" method that can be used to efficiently calculate global illumination effects. However, for realistic images, also these methods require an enormous amount of computing time. That is why Ulaf Schmidt from the University of Paderborn is looking into parallel implementations. Then the hpcLine can be used as a large render server, that can be accessed remotely by people who want to need to render images. A new development is to apply load-balancing strategies to these parallel implementations.