Contract with IBM of the German High-Performance Computer Center North (HLRN) is partly signed

In a competition situation, the German Science Council decided that Leibnizrechenzentrum Munich should install The German Supercomputer in Summer 2000 instead of HLRN. That contract was signed end of October 1999. Thus Northern Germany had to wait.

The procurement process started beginning last year and is now finished. On March 18, the Minister of Science and Culture of Lower Saxony, Thomas Oppermann, and the German IBM CEO, Erwin Staudt, signed the contract at CeBIT Fair. The Berlin partner will also sign it this week. It is extremely remarkable that in a financial critical situation of the Federal States and Germany the sum of 20 Million Euro will be invested for scientific computing. Ten Million Euro are paid by the North Germany states Berlin, Bremen, Hamburg, Mecklenburg-Vorpommern, Niedersachsen (Lower Saxony) and Schleswig-Holstein, the other ten are paid by the Federal Republic Germany because of the Hochschulbauf??rderungsgesetz (law for funding buildings and investments in the universities). An other extremely important aspect is the deletion of the federal boundaries. As a researcher at the University Hannover of Lower Saxony one legally - from the federal viewpoint - had to pay real money for the usage of a computer in an other state. But often researchers and the supercomputer centres found alternative ways. In 1984 the three states Berlin, Niedersachsen and Schleswig-Holstein started the North German Computer interconnect. The centres shared their computing resources. Thus the researchers could use the best machine at that time in one of the three states.

At the locations Berlin, Zuse-Centre Berlin (ZIB), and Hanover, Regional Computer Centre of Lower Saxony (RRZN), IBM installs a massively parallel computer of the pSeries 690. Each site gets 12 compute nodes. A node consists of 32 IBM Power4 processors, 1.3 GHz. This sums up to 384 processors at each site and 768 processors in total. A Power4 processor produces 4 floating point results per clock, 5.2 GFlop/s. This leads to a peak of nearly 2 TFlop/s at each site or 4 TFlop/s for the total HLRN computer. The total memory sums up to 2 TByte. IBM will start the installation in the first half of 2002. Following the IBM results in the TOP500 list it can be expected that this machine will run with an Rmax of 2.4 to 2.7 TFlop/s and then will be within the Top10 or Top20 of the next list.

The computer/computers can be accessed from universities and scientific institutions of the participating states using the German scientific network of the DFN (German Research Network), the WiN.

The future-directed co-operation of the states in Northern Germany will influence the regional infrastructure in science and industry.

Four Levels of Parallelisation

The lowest level is based on the Power4 processor with 4 floating point results. The next level 32 Power4 processors are connected to an SMP-node with a shared memory of 64 to 256 GByte. The peak performance of such a node lies in the range of 166 GFlop/s. The third level are the computer centers of RRZN and ZIB with 12 nodes each which are internally connected by the high-speed IBM switch to a computer with 2 TFlop/s. The fourth level is a grand challenge for both computer centres. The computer centres will be coupled by a fibre channel connection with a bandwidth of 2 Gbit/s. From a users' perspective the total system, including the 300 km network connection, is a distributed, homogeneous massively-parallel system with a peak performance of 4 TFlop/s and 2 TByte main memory. Thus a job, who uses the 24 nodes, has to communicate between the two computers via the fibre channel. The realisation of the "One-System Property" is a specific technical challenge for HLRN as well as for IBM.

Application Areas

The states analysed a demand for high-performance computing for some specific Northern German applications. Research activities lie in the science areas of ship building, coastal and sea as well as climate research, biological waste water techniques, ozone and noise research as well as different engineering applications. It will be used for basic research in physics, chemistry and life science too.

The old computers

The users of RRZN access an ?older" Fujitsu Siemens vector computer VPP300 with 4 processors. The vector programs are ported on clustered systems, e.g. the Sun E10000, which has such an architecture. The Cray T3E at RRZN has a small number of processors and was used for program development for the big Cray T3E at ZIB in Berlin with 512 processors.

History of RRZN and ZIB

The Regional Computer Center of Lower Saxony in Hanover was founded in 1973 and started with the good old Control Data Corporation Cyber 73 and as the supercomputer the Cyber 76 - the fastest computer of that time. It served users for high-performance applications in Lower Saxony and some research institutes outside Lower Saxony in Heidelberg, Hamburg and Kiel.

The Zuse Institute Berlin (ZIB) is a research institution out of the Berlin universities. In co-operation with the universities research and development in the field of computer-science in application-oriented algorithmic mathematics and practical computer science. Additionally it offers high-performance capacity as a service provider. It was founded in 1984 and installed a Cray 1M, which was used by Berlin, Lower Saxony and Schleswig-Holstein university researchers.

Personal Comment

I want to congratulate both, RRZN and ZIB, for this procurement. It will lead them into the peak HPC computer centres in Germany and in Europe. I started at RRZN in the beginning, 1973, and was co-ordinator and in the advisory service for the ZIB Cray users in Lower Saxony until 1985. Thus there is a very personal connection with both centres.