Expert group for fusion computation established
Specialists at IPP supervising fast computer codes / support for research throughout Europe
An expert group for high-level computations in fusion research will start work at Max Planck Institute for Plasma Physics (IPP) in Garching, Germany with the opening ceremony on 20 May 2009. Partly funded by the European Union, this High-Level Support Team is being established at IPP to support scientists in the European Fusion Programme in order to make their computation codes suitable for processing on supercomputers.
The objective of fusion research is to derive energy, like the sun, from fusion of atomic nuclei. To ignite the fusion fire, one has to confine the fuel – a very low-density, ionised hydrogen gas, called plasma – in magnetic fields, keep it stable and thermally insulated, and heat it to temperatures of over 100 million degrees. The complex behaviour of the plasma is being experimentally investigated in numerous fusion devices word-wide. The next major step is the ITER (Latin for “the way”) international fusion test reactor, now being built at Cadarache, France. But in the computational description of fusion plasmas there have also been notable achievements. The long-term goal is complete computer simulation of the processes in a fusion plasma, e.g. turbulent motion in the plasma, processes in the plasma edge, and the effect of various heating methods.
Besides physics know-how, such simulations call for fast high-performance computers: the HPC-FF (High-Performance Computer for Fusion) with a capacity of 100 teraflop/s, i.e. about a hundred trillion computation steps per second, will soon be available at Jülich Research Centre for fusion scientists throughout Europe. A good ten times faster petaflop computer is to be acquired as of 2012 for the International Fusion Energy Research Center (IFERC) at Rokkasho as a Japanese-European cooperation. The faster computers can process large quantities of data, the more sophisticated are the problems that can be solved: a petaflop computer should allow simulations over the full 800 cubic metre volume of the ITER plasma.
To manage 100 trillion computation steps in one second – this would take a few million years with a pocket calculator – modern supercomputers use ten of thousands of processors simultaneously. But for such „massively parallel systems“ the computation codes of plasma theoreticians are not immediately suitable. To ensure that the enormous capacity of the computers is properly utilised, the specialists at Garching have therefore to apply sophisticated mathematical methods to tailor the programming exactly to the allocated computing capacity. “Five members of the expert group will work at IPP and Garching Computer Centre, both of which have acquired many years of expertise in the field of computational physics. The other four members are at home in other European fusion laboratories”, states Prof. Dr. Sibylle Günter, Head of the Tokamak Theory Division at IPP and Chairman of the HPC Council.
One of the purposes of the HPC Council is to ensure optimum utilisation of the Jülich computer: “With HPC-FF at Jülich and the expert group at IPP in Garching European fusion research is well equipped with the tools and know-how needed for effective preparation and evaluation of the ITER experiments. And we are now able to produce numerical models required to design a demonstration power plant”, Prof. Günter goes on to state. A further objective is to make European fusion theoreticians fit for the coming computer generation, viz. petaflop computers: “This means that we have to start now with the training of scientists and further development of computation codes.”
The opening ceremony at IPP on 20 May will include a presentation of the first theory projects which the HPC Council has selected for optimisation from numerous applications submitted from the whole of Europe. For four years the HPC project will be funded in the framework of the European Fusion Development Agreement (EFDA), then it will be decided whether to grant an extension.