INDUSTRY

By Wolfgang Hafemann, IWR-University of Heidelberg -- The Interdisciplinary Centre for Scientific Computing of the University of Heidelberg (IWR, http://www.iwr.uni-heidelberg.de) has just finished the installation of a parallel high-speed PC cluster (http://www.helics.de) with an excellent, yet unknown price/performance ratio. This parallel high-performance computer (named HELICS), which has been acquired together with the computing centers of the University of Mannheim (RUM) and the University of Heidelberg (URZ), is set up with standard components (so-called ‘Commodity off the Shelf’ components). HELICS consists of 512 AMD Athlon MP PC processors; two of them are placed into one computing node. These processors have frequencies of 1.4 GHz and reach a theoretical peak performance of 2.4 billion floating point operations per second (Gflops) for each processor. The interconnections of the computing nodes are realized via Myrinet 2000, a high-speed network from Myricom (http://www.myri.com), one of the current leaders in high performance networking. This network put together as a CLOS network has a bisection bandwidth of 128 x 2 Gbit/s and indicates bi-directional point-to-point throughput of 2 Gigabit per second. In comparison to conventional networks, e.g. Ethernet, Myrinet2000 allows better scaling behavior of parallel applications. The system is equipped with a total of 256 GigaBytes of distributed memory. The file storage reaches approximately 10 TeraBytes. The total system achieves a theoretical peak performance of more than 1.4 Teraflops, which even exceeds the performance of all other presently known Myrinet PC cluster systems installed in the USA. Since 1993 the University of Mannheim, together with the University of Tennessee, has introduced a list (‘Top500’) of the 500 fastest computers of the world (http://www.top500.org). This list is known to be an important instrument for the evaluation of the efficiency of high-performance computers and is accepted world-wide. The positioning in the Top500-List is related to the well-known Linpack benchmark. First performance measurements of HELICS have indicated a Linpack performance of exceptional 825 Gflops. In the latest Top500-List of November 2001, the system would have reached position 24 worldwide, position 4 European-wide and position 3 Germany-wide. The costs of the HELICS Cluster Computer were approximately 1.26 millions EURO (1 Million USD), a comparatively small amount of money compared with what usually must be spent to achieve such high performance, typically known only from supercomputers. The HELICS Cluster therefore has one of the worldwide best price performance ratios. This is possible because of the consistent use of low-priced standard components available at the market as well as of the cluster computing knowledge at the IWR. The system was supplied by the German company Megware, who already installed the well-known PC cluster at the University of Chemnitz. The project is supported additionally by AMD (Advanced Micro Devices), which succeeds to position their hardware solutions in the area of High Performance and Supercomputing. The Interdisciplinary Center for Scientific Computing (IWR) is a central institution of the University of Heidelberg for research and teaching in the area of scientific computing. At the IWR, different groups have been put together to cooperate in developing methods of mathematical modeling, based on the use of parallel computers. The interdisciplinary cooperation of researchers of different fields, the development of efficient computing algorithms and extremely efficient computer equipment are crucial prerequisites for the successful work of the center. The parallel computer will be used by groups of researchers at the IWR and the University of Mannheim in particular in the disciplines of mathematics, computer science, bio-computing and chemistry as well as physics for extensive calculations and simulations (‘Grand Challenge Problems’). Since 1989 the IWR operates as a development center for parallel software applications within the diverse areas. Key points of the current research and technical applications are the areas of: • Reactive flows, • Optimization problems, • Technical simulations, • Image processing, • Bio-computing/Bioinformatics. In particular, bio-computing has a special and increasing meaning at the IWR, since this area takes a central function in the ‘Rhine bio region’ and the ‘Heidelberger Genomforschung’. The research is focused on the questions of efficient use of parallel hardware at the IWR (adequate parallelism, efficiency, scalability and portability of algorithm). The high-speed parallel computer at the IWR becomes a substantial instrument for its research. The expertise in parallel computing at the IWR is also used by other groups of researchers and research institutions. With its parallel computer the IWR represents also a center for national groups of users (key programs of the DFG, national and international research centers for molecular biology and cancer research, industrial partners). The participation of groups of researchers in the IWR and the use of its resources are based on scientific projects of the members. But also for external users, e.g. industrial partners, HELICS will be available in the context of the HWW co-operation (‘High Performance Computing Center for Science and Industry’). Parallel computing experiences a steadily increasing boom in the area of the scientific and commercial computing over the recent years. Scientists take part in the development of clusters for parallel computing, their usability and integration and the parallelization of applications in general. For the development of software and parallel algorithms with the necessary tests and investigations of the run time behavior a direct access to parallel systems is required. The operation system of the parallel computer makes an interactive operating of the parallel hardware and software possible. The basis for this direction of the PC cluster computing is the well-known operating system Linux, the open PC Unix system, which was developed by Linus Torvalds and has become popular world-wide. Additionally, HELICS system uses Score, a software environment developed by the RWCP project in Japan. The further necessary software tools originate from the area of the worldwide ‘Linux Community’ and are adapted to the special needs of the supercomputer. With its parallel cluster HELICS, Heidelberg will surely take a leading position in the area of the ‘High Performance Clustering’. With a scalable cluster computer like HELICS, the IWR will reach the area of the Teracomputing.