Features:

PARIS-LODRON UNIVERSITY SALZBURG INSTALLS SCI-CLUSTER

Gaisberg is the name of an eye-catching mountain on the outskirts of Salzburg with a radio broadcast station on top of it. Now, it is also the name of the new super-computer which was installed in late 2003 at the department of Scientific Computing at the Paris-Lodron University Salzburg. It is a cluster of 25 computing nodes with 2 Athlon processors each, connected by an SCI 2-D torus network. Given the requirements of scientific computing, such as large data sets, time-demanding algorithms, and the flexibility to support a variety of applications, programming languages and scientific software-libraries, the Linux cluster offers a very good solution at a reasonable cost-performance ratio.

The cluster was assembled and installed by MEGWARE Computer GmbH based in Chemnitz, Germany. Megware specialises in building, integrating and supporting Linux clusters. More than 100 clusters have been installed in renowned institutions within Europe. The two-processor computing nodes come in special CoolNode cases that are Megware's own design and guarantee optimal cooling of the 50 AMD AthlonTM MP2800+ processors.

SCI (Scalable Coherent Interface) PCI interface boards based on Dolphin Interconnect Solutions's technology are used for the high speed network connections. The network forms a two dimensional 5x5 torus. An SCI network was chosen for its extremely low latency and high bandwidth. With 50 gigabytes of total main memory and over 1.5 terabytes of total hard disc capacity, the cluster is capable of serving as high-bandwidth database system. The cluster management and message passing software (MPI)has been developed by Scali AS, an Oslo based software company specializing in clustering solutions.

High Performance Computing has been a tradition at Salzburg University since the newly established department of Computer Science in a joint effort with the department of Mathematics started to solve big computational problems on a Convex. The new cluster offers a computational power that is about 16 times greater than that of the SGI Powerchallenge with 20 processors used so far.

At the department of Scientific Computing the machine will be used for parallel linear algebra in the field of intelligent search engines and text- retrieval. Matrix decompositions such as the singular value decomposition (SVD) are able to transform the word-space of text documents into a concept- space, which improves search results of semantic document searches significantly by automatically considering words with equal meaning as possible search hits. However, these matrix operations are computationally expensive. The demand for parallel algorithms and high performance computing facilities is obvious.

Another application at the department is parallel multimedia processing. As nowadays the flood of digital mutlimedia content is constantly increasing, the search for efficient compression algorithms leads to modernized but also more resource demanding compression standards. In case of large data sets and/or realtime requirements, parallelized algorithms are an indispensable part of multimedia applications.

The Salzburg cluster will also be used by other research groups at the university for applications in mathematics, chemistry, geography, physiolocial psychology and mineralogy. It is also planned to make it available to other academic institutions and local industry. In addition, the system will be integrated as a grid node in the Austrian Grid and the EU-wide CrossGrid.