INDUSTRY

By Uwe Harms, Harms Supercomputing Consulting -- Professor Friedel Hoßfeld headed ZAM (Central Institute for Applied Mathematics) at Research Centre Jüelich (FZJ) for nearly 30 years. During this time, he influenced the German supercomputing scene dramatically and made the ZAM Computer Centre one of the leading centres in Germany, Europe, and the world. After operating big IBM computers in the 70s, he started with the first vector supercomputer in 1983, a Cray X-MP/2, the first worldwide for civil applications. Then he moved through the Cray vector series with another Cray X-MP/4, via a Cray Y-MP up to T90. The last big installation took place in 1996/1997 with two Cray T3Es and a 16-CPU Cray T90. This led ZAM to rank 24 in the November 1996 Top500 list and on rank 10 and 19 in the June 1997 list, when all these computers had been installed and upgraded. The replacement of the T3Es is scheduled for next year. He answered questions concerning the worth of expenditure of tens of millions of Euros for supercomputers, the technology expectations of the next 20 years and also personal wishes to the German Ministry of Education and Research. FZJ and ZAM organized a ceremonial retirement colloquium on July 11, 2002. There,. I interviewed Professor Hoßfeld. He heavily influenced the German supercomputing scene by tight cooperations and interactions between the academic and research centres. Additionally, he heavily pushing the ideas of computational science and engineering (CSE) in the Ministry of Education and Research as well as in the scientific community. The challenges of CSE need the support by supercomputers. In addition, he actively participated in a diversity of strategic committees. He is the representative of the Hermann von Helmholtz Society of German Research Centres (HGF) in the National Coordination Board for the Procurement and Usage of High Performance Computers of the German Science Council (Wissenschaftsrat) who established this board according to the “Recommendations for the future usage of High-Performance Computers” which the Science Council issued on May 12, 2000 (Drs. 4558/00). Professor Friedel Hoßfeld studied Physics in Würzburg. At the Max-Planck Institute in Würzburg where he dealt with Fourier-Transforms to investigate crystal structures, he got to use a Zuse Z22 R computer. In 1964, he changed to FZJ and became head of ZAM in 1973. In 1983, FZJ acquired the first supercomputer, a Cray XMP/2 – the first in the world which was dedicated to civil research. In 1987, he built up the first German Supercomputer Centre (Höchstleistungsrechenzentrum, HLRZ). This was a joint venture of DESY (German Electron Synchroton) in Hamburg, GMD (German Research Centre for Mathematics and Data Processing) in St. Augustin, and FZJ which opened its resources for qualified scientists mainly in physics and chemistry from all over Germany. This was a big step forward, as then the federal structure of Germany in that respect was a barrier for the access to computers from one county to another. HLRZ was considered by the Science Council as the prototype when he recommended, in summer 1995, to install up to 4 German Supercomputer Centres. Researchers in Germany can use theses resources for free, if they got the approval of grants by independent scientific boards and steering committees after having submitted research proposals in well-founded projects for peer-reviewing. This procedure, too, was implemented for first time at HLRZ. When GMD left the joint venture in 1998, FZJ and DESY founded the John-von-Neumann Institute for Computing (NIC). Harms: Do you think it is worth it to invest 20 or 50 million Euro for supercomputing? Hoßfeld: What particle accelerators mean to experimental physics, supercomputers deliver for simulation – the third category of scientific research besides experiment and theory: supercomputers are the „Accelerators” of the theory. As large-scale instruments for the scientific infrastructure, they are subject to the same criteria as big experimental facilities. The question of return on investments in science and research is difficult to answer, as usually. Economic measures only apply to the efficiency of operating these systems, but generally not to the research projects run on them. Therefore, in all supercomputer centres the principles of granting and evaluating research projects have been implemented which have been applied and approved in peer-reviewing by the Deutsche Forschungsgemeinschaft since decades. The steering committees and scientific councils are heavily working and meanwhile quite experienced in these procedures. Thus the issue is not the financial return on investment but the international competitiveness of German science and research. The USA with their ASCI program of DoE and the PACI initiative of NSF and, just recently, Japan with its “Earth Simulator” push the innovation in scientific computing far ahead that is it difficult for us in Germany and in whole Europe to keep pace; I am afraid that we will hopelessly lag behind – despite all respective efforts, we hopelessly lag behind – despite of all respective efforts. Harms: ZAM operated for nearly 20 years the fastest computers of their time. What do you expect in the next 20 years? Hoßfeld: The evolution of computer performance follows since a quite some time closely to Moore’s Law which means exponential growth. The breakthrough in parallel computing, especially the recent SMP-node based systems, has forced the performance curve far above the TeraFlops level – as the June 2002 Top500 list shows. One of the driving forces have been the US ASCI programme: This spring, the Japanese “Earth Simulator” jumped up to more than 40 TeraFlops. It dominates the June 2002 list and will further trigger the development of supercomputers. The resulting “Computenik Shock” in the US will certainly lead to strengthened efforts and financial support of supercomputing developments there in order to gain back the „technical leadership”. Definitely, the Earth Simulator is an interesting mixture of the SMP-structural concept and NEC’s vector computer architecture and technology. The chip technology used in scalar SMP systems in combination with air cooling instead of liquid cooling challenges air-condition techniques and energy supply beyond present infrastructures everywhere. Additionally, the needed floor space reaches thousands of square meters, if performance far beyond the TeraFlops border is required. New computer rooms, which primarily have to fulfill the cooling requirements and to offer enough floor space, are the consequences. This status of computer architecture corresponds to today's technological and technical capabilities to achieve such high computer performance. The further growth towards PetaFlops on the basis of the present and predictable technologies of commodity chips and components will not provide a reasonably viable roadmap. New technologies are necessary. The PiM (Processor in Memory) technology, which is explored especially by IBM, could be a promising way. However, the question arises whether well-balanced systems can be built with this technology. At the end of this decade we will trivially have a clearer view. The limits of growth out of Moore’s Law are foreseeable. If one inverts Moore’s Law by asking respectively for the number of atoms which are available over the years to represent a bit, the decaying exponential curve cuts the level „1” around the year 2017. Then, irrevocably, the quantum regime will take over. It is impossible for me in this moment to forecast the technical feasibility of the quantum computer; yet the impressive reports published almost weekly on experimental achievements in this field are promising. However, I am convinced that in Germany a strong impact on research and development to explore the potential and to initiate the design of the quantum computer and quantum algorithms is more than overdue. Unfortunately, the drastic cut of our headcount at ZAM in the 90s did not allow to enter in time the research field of quantum computing, and I mean that half-heartedness is not sufficient also in this case. Harms: If you would have a free wish to address to the German Ministry of Education and Research (BMBF) concerning supercomputing – what would you say? Hoßfeld: As this question belongs to the area of fairy tales, I first would wish a good fairy in the Ministry. Then I magically would have three wishes free. With the first wish, I would wish in our Ministry more of the high command and “Spirit of Technical Leadership” of the USA in our supercomputing field. In the second, I would ask the good fairy to undo the drastic cuts of ZAM staff by 23 jobs and of the financial resources by about a third within the last ten years. Then we could stabilize our competence in scientific high performance computing in ZAM. Further on, we could attack more innovative topics to enhance and exploit future computing, which naturally have a home in ZAM – in my view – compatible with the spirit of John von Neumann. Especially, quantum computing would be an adequate field of research and development. The research work at ZAM will be an essential contribution to the Research Programme “Scientific Computing” in the Research Area “Key Technologies” of HGF. And finally, I would keep my third wish for future “accidents”, if again someone would have the strange idea to shorten the resources of ZAM. Harms: When will ZAM get new supercomputers, to play again in the first supercomputing league? Hoßfeld: In the last three years, we worked very hard on the procurement of a new supercomputer. Our efforts have been strongly supported by the Board of Directors of FZJ and the Committees of FZJ and NIC as well. We will install a pretty large IBM SMP-system with a peak performance of about 6 TeraFlops. It will be installed in 2003 as soon as our new building at ZAM will be completed. For more information visit www.fz-juelich.de/zam or www-fz-juelich.de/nic.