So how are the German Supercomputer centres doing?

In Germany supercomputers have been installed in regional centres all over the country. Today major investments have been done by the Federal Republic and the counties in this area in both research and academia. The traditional industry sector fell back, but some automotive companies have installed new systems this year or are involved in a cooperation with academia. In this background report several of these centres used computers and the tasks are summarised. In particular, the cooperation of academia and industry is discussed.

Konrad-Zuse-Center for Informations Technology (ZIB) in Berlin

The first machine was a Cray 1M in 1984 which was followed by some other powerful Cray vector systems. In 1995 they installed a Cray T3D with 256 processors, which needed a Cray Y-MP 4 processors as a host, and had a pool role for several years. It was complemented by a Cray T3E in 1997 with 152 (136 for parallel applications) processors with a peak performance of more than 135 GFlop/s. By the end of this year (1998) the T3D will be shut down and probably( mid November) the T3E will be extended by 72 processors (DEC Alpha EV5, memory 129 MByte, peak performance 900 MFlop/s). Then the system will have 208 processors for parallel applications. After the 30 days acceptance phase the T3D will be disassembled. The I/O and the disc space of the T3E will be extended. It should be mentioned that there is a Cray J90 with 32 processors too. Research Centre Juelich

In summer 1996 the "really first" German Supercomputer Centre was opened in Juelich. It consists of Research Centre's computer centre and of HLRZ (High-Performance Computer Centre for theoretical Physics/Chemistry), which was a joint project of Juelich, DESY (German Electron Synchrotron) and GMD. It is equipped with a Cray T3E with 512 processors (more than 300 GFlop/s peak), a 1997 Cray T3E with 256 processors (230 GFlop/s peak), a T90 16 processors 29 GFlop/s peak) and a Cray J90 with 20 processors. In total the performance sums up to more than 560 GFlop/s. This underlines that Juelich is one of the major supercomputer centres in the world. The Rmax, the Linpack-performance, is about 20% of Germany. An important fact is the distribution of the usage, 15% is dedicated to industrial projects - chemistry/material sciences/bioinformatics are important areas, where Juelich has profound experiences in its institutes, 50% HLRZ can use and 35% are for their own needs. Other industrial topics are services in exploration and environmental applications.

RUS/HWW (Hoechstleistungsrechner fuer Wissenschaft und Wirtschaft Betriebsgesellschaft mbH = HPC for Science and Industry Operation Enterprise)

Behind this terrible word one finds an innovative approach, academia and industry combine their strength in HPC and can use different architectures. The partners hold the following shares, County Baden-Wuerttemberg 16.8%, University Karlsruhe and University Stuttgart 16.6% each, debis Systemhaus (Daimler-Benz Inter Services) 40% and Porsche AG 10%. Debis Systemhaus sells computing time and services on these machines especially to small and medium enterprises in Germany and outside and delivers services and computing time to the Daimler-Benz group too. This constellation allows new perspectives of the knowledge transfer from science into industry. The institutionalised exchange of experiences between research and industry will improve the transfer of information from academia to industry, thus securing the position of Germany.

In the first step only the University of Stuttgart was involved, but now the machines of the University Karlsruhe are integrated (see Virtuelles Supercomputercenter Karlsruhe). In HWW the users have access to a Cray T3E 512 processors with a peak performance of 460 GFlop/s, a NEC SX-4 32 processors (8 GByte SSRAM, 32 GByte DRAM) with a peak of 64 GFlop/s. The partners from the industry integrate their own machines, Cray C90, T90-4, J90-8, IBM SP2. In Spring this year, an additional NEC SX-4 4 processors and 8 GByte SDRAM was bought for applications with big memory requirements like CFD and aerospace applications. As development platforms are an Intel Paragon XP/S till end of 1999 and a Hitachi SR2001 available. The Cray architectures are concentrated in Untertuerkheim at debis Systemhaus while the others are located in Stuttgart-Vaihingen at RUS (Univ. Stuttgart Computer Center).

As this centre is listed as academia in the Top500 list, the usage of the industry is hidden. Contractually about 7% of the total capacity is reserved for industry, that mean about 37 GFlop/s counting the peak performance. Thus different architectures are available for the industry as well as for academia. This centre is open too for other industrial users, for xample small and medium enterprises, via debis Systemhaus. At present the NECs are used heavily from industry, automotive and aerospace in the Daimler-Benz group.

Virtual Supercomputer Center Karlsruhe

In Karlsruhe too academia and research have bundled their resources, but only the machines of the university belong to the HWW. Now the University Computer Center has an IBM SP2 with 256 processors and the huge memory of 122 GByte, the peak performance sums up to 68 GFlop/s. At the Research Center Karlsruhe a vector architecture is installed, a Siemens-Nixdorf/Fujitsu VPP300 with 16 processors and 32 GByte distributed memory, 35 GFlop/s. 28 SP2-processors are owned by the Research Center and 12 of the VPP300 processors, the rest is for the university.

Max-Planck-Society (MPS)

MPS has ordered a lot of Cray T3Es. The biggest machine is installed in the Institute of Plasmaphysics in Garching near Munich. In February this year they extended the 672 processor system to 784 application processors. As they use "old" 600 MFlop/s Alphas, the peak performance sums up to 470 GFlop/s. Other institutes of MPS use 16 processor T3Es at their location, the Institute of solid materials ind Stuttgart, Haber-Institute in Berlin, Institute of Polymers research in Mainz and Institute for Fluid Research in Goettingen and are connected with the central computer.

Leibnizrechenzentrum (LRZ) in Munich

LRZ extended its VPP300 from 32 processors in the beginning of 1998 to the contractual number of 52 although it was planned for summer. Now it has 104 GByte of memory and 114 GFlop/s peak performance. The installation of the system was supercomputer-like. Purchase end of December 1996, delivery in April 1997, start of the acceptance test on May 26, end of acceptance test June 24 with a 100% availability. The good experiences - like MPS - with the same machines at the remote universities, in former times Cray, Bavaria followed this line. The University Nuermberg/Erlangen bought a 6 processor VPP300, the universities Bayreuth and Wuerzburg ordered a VPP300 with 2 processors and the University Regensburg a VX/2 (2 processors), the "office" model of the VPP300. Additionally LRZ uses an IBM SP2 with 77 processors, 20 GFlop/s peak.

The Bavarian state wants to extend LRZ to a German supercomputer centre, like Stuttgart. The offered financial support. Thus LRZ is driving the ideas and hopes to realise the first step around the year 2000. Here too a cooperation with industry is a target.

Eastern Germany - the new counties

In the eastern part of Germany HPC has found its way. This year six systems are listed in the Top500. The leading machine is a Hewlett-Packard X-Class with 64 processors, used for image processing in industry. All the other machines are installed in academic and research institutes. HP has sold well there, two other X-Classes at the universities in Leipzig (48 processors) and Halle-Wittenberg (32 processors). SGI/Cray is represented at the university in Rostock with a Cray T3E with 50 processors and a SGI Origin 2000 with 56 processors in Dresden. IBM sold an SP2 with 77 processors to PIK ( Potsdam-Institut fuer Klimafolgenforschung, research on climate consequences) in Potsdam.

Industry

Usually HPC is used in the technical-scientific market, represented by automotive, aerospace, chemistry and exploration for example. But this picture is wrong in Germany, in the Top500 list only Volkswagen uses its NEC SX-3 with 3 processors for automotive applications. They have some more new NEC SX-4 computers, but each with a number of processors that lies below the entrance performance of the list. It has to be mentioned that Daimler-Benz and Porsche access the HWW computers. Ford with its Cray C90 with 14 processors has left the list. Two other machines have to be mentioned in this context, BMW, Munich, acquired a Cray T90 with 12 processors - the former Juelich machine - in spring this year and Audi, Ingolstadt, bought a Siemens-Nixdorf/Fujitsu VPP300E with 16 processors that should be installed end of June this year - just now. Mostly industry is using minor HPC-systems like Origins or PowerChallenges that have not the performance to enter the list.

In Germany one can observe that not the "classical" applications use HPC systems, but finance, telecommunications and database. Out of the 17 industry HPC machines 15 are used in these areas. Surprisingly 14 of these computers are Starfires, SUN HPC 10000. Remarkable in the Commerzbank with 7 SUNs. Their aggregate Linpack-performance sums up to 216 GFlop/s and is nearly the same as that of the Cray T3E-512 at the Research Center Juelich, rank 17 of the Top500.

The history of German Supercomputer Centers

Very early academia and research have been interested in supercomputers with the performance of that time. All these "old" centres are still on the leading edge or trying to get new systems. The Computer Centre of the University of Stuttgart (RUS) bought a CDC 6600, the RRZN (Regional Computer Center of Lower Saxony at University Hannover) installed an CDC 7600 in 1973, the Regional Computer Center of Cologne had the same machine, Max-Planck-Society bought a Cray 1 in 1979 for the institute of Plasmaphysics (MPI). The computer center in Bochum bought a Cyber 205 in 1981, followed by the Research Center Juelich, the DLR and RUS with a Cray in 1983, 1984 ZIB in Berlin followed. Nearly all these centers have underpinned their importance as a regional or national center till today.

In 1970 DFG (German Research Society) developed the concept of the regional centers, resulting in the centers in Berlin (ZIB), Hannover (RRZN), Cologne, Munich, Stuttgart(RUS)/Karlsruhe and some others. Three years ago, the Scientific Council decided that Germany should fund only two to four German supercomputer centres. The first results can be seen today, in the research area the centers in Juelich and MPI in Garching, in academia RUS in Stuttgart, started as the first. More centres as Leibniz Rechenzentrum (LRZ) in Munich and a solution for Northern Germany, distributed between Berlin (ZIB) with an MPP system and Hannover (RRZN) with a vector processor and graphics/visualization will follow in the timeframe 1999 to 2001. In northern Germany several counties, actually 6, are cooperating and pushing this project. These centres will operate high-performance systems and offer services for researchers all over Germany. Usually a researcher in academia has to pay real money if he uses a computer in an other county. As early as in 1984 Berlin, Lower Saxony and Schleswig-Holstein agreed to "pay" in computing time, this resulted in a computing cooperation, the first in Germany.

An other important aspect in these centers is the industry orientation. Especially SMEs (Small and Medium Enterprises) can use such systems, having access to advisory services. In Stuttgart such an approach is now possible.