Supercomputers

At the end of May 2009, the Forschungszentrum Jülich proudly presented its three new supercomputers: JUGENE, JUROPA and HPC-FF. With a computing capacity of 1 petaflop – equal to 1 quadrillion computational operations per second – Jugene ranks third in the league table of the world’s hundred fastest computers as published in June. The Tandem Juropa – HPC-FF still holds tenth place with 308 teraflops – or 308 trillion computational operations per second. However, the race for computing power is not an end in itself, says Prof. Dr. Dr. Thomas Lippert, Director of the Jülich Supercomputing Centre (JSC) at the Forschungszentrum Jülich. “Meanwhile, supercomputers are, besides theory and experiment, the third pillar on which research stands, and their efficiency and performance capacity create the prerequisites for solving extremely complex questions.”

Where real-life experiments are too dangerous, such as in safety assessments of nuclear power stations, are too complicated and expensive, such as in the case of the fusion reactor ITER, or are even impossible, such as in studying the causes and consequences of climate change, computers serve as virtual labs today. Using simulations, researchers try to replicate reality and predict future behaviour. “Supercomputers are the only option for solving many scientific problems at all,” says Lippert. They can calculate much faster, because they not only run on one processor, but rather hundreds of thousands run in parallel. 290,000 processors lie in Jugene. A process that would preoccupy a normal PC over ten years can be completed here in 20 minutes, making it possible to repeat with various parameters.

However, supercomputing is also a science in itself. On the one hand, it calls for special technology. It is – in the case of Jugene–accommodated in 72 telephone kiosk sized cabinets. The confusing mass of countless processors is linked together via a very fast network through which the data are exchanged. “However, not alone the computing power is decisive,” emphasises Lippert, “you also have to know how to use it.” When a programme runs on many processors, it is like a major building site. It’s all about distributing the tasks properly, not all of them can be done simultaneously. Which processor receives new data when and where, how these arrive in good time and reliably, all this has to be cleverly organised and programmed. To support the specialist scientists from various subject areas in optimising their simulations for the respective supercomputer, four dedicated simulation labs were set up at the JSC: for plasma research, for molecular systems, for computational biology, and for atmospheric and climate research. Teams work here that have both the corresponding specialist expertise as well as being experts in supercomputing itself.

The simulation labs and other facilities at the JSC not only offer services for the users of supercomputers. They also carry out their own research and development work. New software tools, for example, that analyse how the simulation programmes could be further optimised. The development of new computers is also on the agenda. Because supercomputers are oneoffs whose computer architecture meets the special requirements of research. The computer clusters Juropa and HPC-FF were designed in Jülich, for example, and were then turned into reality together with companies like Intel, Bull, SUN, Melanox and ParTec. The supercomputer HPC-FF exclusively serves Europeanfusion research. For particularly demanding tasks it can be coupled with Juropa.

The users of the supercomputers in Jülich come from universities, from other Helmholtz Centres,from the Max Planck Society, and from industry. With the new computers, the JSC will also become increasingly attractive for European researchers, believes Lippert. “The seemingly record-breaking computing power sends out an important signal, namely that excellent research can be done here.” To coordinate the activities of the individual European countries in the field of supercomputing and to strengthen their position in competition with the United States and Japan, the Partnership for Advanced Computing in Europe” (PRACE) was initiated in 2007. “The aim is to provide users with more tailor-made capacities,” says Dr. Thomas Eickermann from the JSC, Project Manager for PRACE. In Germany, the supercomputing activities have already been pooled by merging the three supercomputing centres in Jülich, Stuttgart and Garching to form the Gauss Centre for Supercomputing.