Cluster Computing:

J. KEPLER UNIVERSITY LINZ INSTALLS DANUBE COMPUTING CLUSTER

With the Danube Cluster, another computing cluster for high performance tasks went operational at GUP Grid and Cluster Computing Lab (G2C2), which is located at the Johannes Kepler University Linz. Consisting of single PCs and a high-speed network, the system allows researchers to cope with tasks that could not be tackled with a single personal computer. With the resulting price/performance ratio, this solution is especially interesting for organizations with less extensive budgets.

In terms of both equipment and know-how, the Johannes Kepler University is Austria's leading university in the field of high performance computing. The Danube Cluster is the first computing cluster of this type to be built and installed by GUP (Institute of Graphical and Parallel Processing). The system components were supplied by Danube EDV Systeme, the microprocessor manufacturer is AMD, and the high performance networking specialist is Dolphin Interconnect Solutions.

The Danube EDV System is a relatively small company from Aschach a.d. Donau that supplies Linz and it's surrounding area with low priced PC-hardware. Being a small company, Danube EDV Systems met the challenge and proved that small Austrian companies are able to play a role in this market segment. The computing power of the Danube Cluster is delivered by 18 AMD Athlon MP processors, some of which were supplied by AMD for free.

The high speed network connection was established by using the IEEE standardized SCI (Scalable Coherent Interface) based on Dolphin Interconnect Solutions' PCI Boards. The network between the individual nodes is organized as ring (2d torus). The 9 nodes, each of which comprising of two processors, are arranged in a 3x3 grid. The SCI network distinguishes itself by extremely low latency and high bandwidth. Hence, the main memory of the individual nodes can be combined to one single memory space. Thus, each computing node is able to access any desired memory address, regardless of its physical location. (Instead of 1 GByte main memory per node, a total of 9 GBytes of RAM are available.) With more than 1 TeraByte, the Danube Cluster's hard drive capacity is commensurate with its computing power. The system's operating system is based on the Gentoo Linux distribution that has been enhanced to enable communication via the SCI hardware. Additionally cluster management software and an optimized message passing library, developed by Scali, has been installed.

There is a wide range of possible uses for a computational cluster of this kind. Besides traditional applications of high performance computers, the Danube Cluster is used for the research at GUP. This work comprises porting and extending Software, which has been developed by GUP, deploying parallel computer graphics algorithms and virtual reality applications as well as integrating the computing cluster as grid node in the Austrian Grid and the EU-wide CrossGrid.

The planning at GUP, however, is already one step ahead: In addition to extending both currently available computational clusters (by adding more standard PCs) other cluster-technology will be integrated at the GUP Grid and Cluster Computing Lab (G2C2). Connecting the individual computational clusters via GigaBit Ethernet is just another step further, that allows to combine and use them as even more powerful high performance computing system. With this experience in cluster computing it is possible to compare existing pc-cluster solutions in the context of a specific application and provide interested companies with the respective know-how.