The University of Stuttgart* was originally established as a school of engineering in 1829. In 1862, four departments (Architecture, Civil Engineering, Mechanical Engineering, and Chemical Engineering) were integrated into the university, establishing the College of Engineering. The university has changed since then, and has now expanded to 14 departments, offering 140 courses with approximately 21,000 students. In addition, the university has the largest scale of research activity on a contract basis among all of the universities in Germany.
The High-Performance Computing Center Stuttgart at the University of Stuttgart, is one of the four largest computer centers in Germany. HLRS (Hoechstleistungsrechenzentrum Stuttgart) has a vector supercomputer, visual computer, and cluster-structured distributed computer. HLRS provides computing resources for universities in Germany, as well as for industrial applications, such as the shared supercomputer system for Debis System Haus GmbH and Porsche AG. Applications include impact analysis simulation required for the safety design of automobile companies in Germany, and computational flow analysis simulation required for the air conditioner design. The High-Performance Computing Center also provides tools and expertise for science and technology computing.
As one of the leading German computing centers, the center has worked with partners from around the world, such as Sandia National Labs, Pittsburgh Supercomputing Center and Manchester Computer Center in the United States, and the Japan Atomic Energy Research Institute and Tsukuba Advanced Computing Center in Japan. The center has a computer facility to carry out a computation of fluid mechanics, particle physics for scientific research, and impact and structural analysis for industrial users. The computer center conducts research of computation and visualization for both structural and non-structural lattices. Various workshops are held at the center to support users of parallel computation and vector technology.
Key Challenges and Issues in the Current Computing Environment
HLRS has major challenges for calculation in the areas of numeral simulation for air turbulence, molecular science, flow analysis of ground water, astrophysics, and particle analysis. As the amount of data increases every year, HLRS experiences the challenge of managing a database larger than a 32-bit actual memory address space. Therefore, it is necessary to use a 64-bit address space. In order to solve this floating-point, large-scale computing problem, it is also necessary to have a high amount of memory cache bandwidth. Solving these challenges requires a parallel algorithm system with a high bandwidth processor, low latency and actual memory address space. It is preferable to have a shared memory type system larger than 16 CPU. This system enables future performance to be improved easily.
HLRS used NEC* Express5800/1160Xa, capable of loading 16 Intel® Itanium® processors. The impressive performance improvements were obtained by processing calculation result data that exceeded 300GB. In the previous system, it was possible to load only 2GB of the large amount of computing result data (exceeding 300GB) from the disk drive to the main storage at a time. After processing, end-processing of the data was required. However, by utilizing the 64-bit addressing and 64GB large main memory capacity, characteristic of the Express5800/1160Xa Itanium-based system, large-scale end-processing is performed all at once. The computing time was reduced to 1/10 of the time required by the previous system. The fundamental calculation code optimized with the Itanium processor is capable of processing at the same speed or 1.5 times faster than the previous system (HP N4000 PA8600 550MHz). The performance improvement is particularly great in the summation computing program.
Comments and Plan Outline
As for the molecular mechanical benchmark, Express5800/1040Xa with the Itanium processor 733 MHz achieved a high performance of 1.52.0 times that of the RISC processor system (per processor). This enabled molecular computing that was not possible to compute on the previous system. Even larger-scale molecular work can now be processed.
Itanium processors enable the use of the 64-bit actual memory address space. It is expected to have dramatic effects through programming efficiency of large data applications. In addition, further Itanium processor family enhancements and performance improvements will drive even better performance. These factors will allow HLRS to continue contributing to society by providing strong computational platforms and increased support to users in need of advanced scientific and technological computing.