Höchstleistungsrechnen in Germany Part 4: Höchstleistungsrechenzentrum Stuttgart and hww GmbHMunich 05 April 2003 In this issue the actual Höchstleistungszentrum Stuttgart (HLRS) and its co-operation with industry via the enterprise hww GmbH will be discussed. As HLRS/hww has old computers it is just in the acquisition phase. In April this year the proposal goes to DFG (German Research Society). Thus the result of the decision is not clear yet.
The High Performance Computing Center Stuttgart (HLRS) supports scientists from research
organisations in the use of leading edge supercomputer technology and its applications. The
mission of HLRS is to provide its users with access to high end systems, tools and expertise to achieve top international positions in their research field.
In addition to providing compute time and services HLRS actively pursues its own research in
the field of computational science and engineering on supercomputers. In a number of national and international projects both with other scientific institutions and with industry HLRS has gained a leading position in the field of parallel and distributed systems and visualisation.
Together with its partners at Sandia National Labs, Pittsburgh Supercomputing Center, Manchester Computing Centre, Japan Atomic Energy Research Institute, Tsukuba Advanced Computing Centre, and National Centre for High Performance Computing HLRS has established a world wide co-operation that acts as a testbed for the development of software in the field of grid computing.
Therefore it is a very specific institution by the industrial connection in the hww GmbH. It is a joint venture, whose member are the State Baden-Württemberg, 12.6% shares; the University Stuttgart, 12.6%; the Universities Karlsruhe and Heidelberg, 12.4% each; T-Systems and T-Systems Solution for Research (SfR), 20% each; and Porsche AG 10%. The T-Systems companies have former been debis Systemhaus. T-Systems SfR is a joint venture of T-Systems, 75%; and DLR (The German Aerospace Research Centre) with 25%.
The Computational EquipmentActually HLRS/hww uses a zoo of computers, located in two computer centres. The vector computers NEC SX-5 cluster with 2x16 CPUs, 4 GFlop/s each - total of 128 GFlop/s peak, 80 GByte SDRAM for parallel and serial applications. The Hitachi SR8000 has 16 nodes with 8 compute CPUs, 1 GFlop/s each. Thus this machine has 128 GFlop/s peak and 128 GByte SDRAM. It is a parallel computer for vectorisable or massively-parallel applications. Both computers are located in Stuttgart Vaihingen.The Cray T3E900 with 512 Alpha processors, 900 MHz, from 1996 ranks #181 in the Top500 list, 341 GFlop/s Rmax with 486 GFlop/s peak. A Cray SV1 has 20 CPUs. 1 GFlop/s peak each, and 16 GByte SDRAM. Both machines are installed in Stuttgart Untertürkheim at T-Systems with a distance of about 20 km between these locations. Additionally there are some scalar SMP servers, a Hewlett-Packard RP8400, 8 CPUs, 24 GFlop/s peak, 8 GB SDRAM, a SGI Origin, 20 CPUs, 10 GByte memory, and an IBM Regatta, 16 Power4 CPUs, 1.1 GHz, 70 GFlop/s peak and 32 GByte SDRAM. A newly extended machine is a Fujitsu Siemens Computers hpcLine Linux Cluster, 96 nodes with 2 Intel Xeon 2.9 GHz CPUs, 1 GByte RDRAM per node. The nodes are connected using the high-speed network Myrinet. Within the hww GmbH construct, the University Karlsruhe offers an IBM SP2 parallel computer with 256 Power3 CPUs, 375 MHz, and a peak performance of 384 GFlop/s. The AMD Cluster at the scientific computer centre in Heidleberg, named Helics and delivered from Megware, belongs to hww too. The AMD cluster uses 512 1.4 GHz processors and delivers 1.43 TeraFlop/s peak performance. With its 825 GFlop/s Linpack performance it ranks number 64 in the actual Top500 list.
HPC-PortalT-Systems offers computing power to the industry via a HPCPORTAL, SfR does the same for research institutions. Within this portal a customer can quickly call for compute power and application software. Thus it is possible for small and medium enterprises to use supercomputers.The accounting of the machines follows the pay-per-use principle. A CPU hour on the Cray SV1 costs for example 18,45 EURO, a resource time hour on the NEC SX-5 with 2048 MB memory lies in the range of 50 EURO. This compares to 1,45 EURO for one node (2 CPUs) with 1 GB memory on the Linux cluster. Additionally one has to pay for the application software usage, it lies in the range of 25 and 10 EURO depending on the package.
Applications at HLRS in StuttgartAs the University has an engineering background, the heaviest load comes from CFD (Computation Fluid Dynamics), which is followed by Physics, Solid State Physics and Chemistry. The industrial users of hww like DaimlerChrysler and Porsche mostly run engineering applications.
The New Computer Centre and the Planning of New HardwareAs most of the machines are rather old, HLRS/hww is just in the acquisition phase. Now for the first time university and industry can plan the new systems. At the location Stuttgart a Höchstleistungsrechner with a peak performance in the range of 15 TeraFlop/s will be installed. The centre expects an application performance of 2 TeraFlop/s. The new HPC system in Karlsruhe will have 3 TeraFlop/s peak performance. Further very important components are the data storage systems and the interconnect of the centres. Both systems shall be connected with a 10 Gbit/s Wide Area Network. The total investment is in the range of 40 to 50 million EURO.An an example of a hybrid solution hww imagine a vector SMP (Symmetric Multiprocessor System) cluster, a IA64 SMP cluster and a PC cluster delivering an aggregated peak performance of 15 TeraFlop/s. All have access to a 100 TeraByte disc space. As a grand challenge the centre views the workbench for scientific and technical computations. It is a workbench for simulations whose components are highly integrated. Different servers, like supercomputers or visualisation servers, store or access common data. Thus the huge amount of simulation results do not have to be moved between different computers.
Competence CentresThe Universities Stuttgart and Karlsruhe will establish a HPC competence centre. Like HLRB these centres offers their customers IT- and scientifically specialised scientists as advisors.But on the other hand the German Ministry of Research cut in the past jobs in this field at Research Centre Jülich and DKRZ. At ZAM. Jülich, in the last 10 years, the ministry cut 23 jobs and one third of the budget.
Uwe Harms [News on Advanced IT] [Calendar] [Analysis] [IT in Medicine] |
 |