Catching-up in supercomputing: must be done at European level, says expert

Climate change research, genetics, material engineering - supercomputing power has become a core factor for success in both science and engineering as it enables researchers to test their theories and models by simulation. Supercomputers, however, are still largely funded and ...

Climate change research, genetics, material engineering - supercomputing power has become a core factor for success in both science and engineering as it enables researchers to test their theories and models by simulation. Supercomputers, however, are still largely funded and operated at national level, separately, and not as a much more powerful grid. The Partnership for Advanced Computing in Europe (PRACE) project is determined to change that by creating a permanent European supercomputing infrastructure.

The European Commission will invest more than €20 million in PRACE under the Seventh Framework Programme (FP7) over the next two years. And by mid-2009 or the end of that year at the latest, the project partners from Austria, Finland, France, Germany, Greece, Italy, the Netherlands, Norway, Poland, Portugal, Spain, Sweden, Switzerland and the UK hope to be able to build a petaflop/s system: a machine that is capable of one quadrillion operations per second.

The vision is the following: establish a grid of supercomputers spread across Europe, allowing the best European scientists and engineers access to supercomputing power that meets the international challenge, a task that national supercomputing centres are not equal to. They would be given access regardless of which country they are from, the criteria being peer-review of their projects.

'Science and economy need processing power on the highest level - the former in order to perform cutting-edge research, the latter in order to create innovation,' said Professor Achim Bachem, chairman of the research centre Jülich and coordinator of PRACE. 'In all natural sciences, supercomputers have become an indispensable tool. In the future, large knowledge leaps will only be made with the help of simulation sciences.'

'I think in the US, they really discovered that supercomputing is the key technology for science or for economic issues,' Professor Bachem explained in an interview. 'They have about 10 times more power than Europe in this key technology. So, the USA is definitely number one, but Japan and Asia are also very strong competitors.' According to the TOP500 project that ranks the 500 most powerful publicly-known computer systems in the world, seven out of the top 10 systems are situated in the USA. Only two of them are in Europe, one in Germany and one in Sweden, while one can be found in India. 'If we want to support our industry, I think we have to catch up and this we can only do at European level,' Professor Bachem pointed out.

This is all the more true for this field since trying to catch up with the leaders in the domain of high-performance computing (HPC) will require considerable investment: 'The infrastructure for supercomputing for the challenges of today and the next years will cost about €400 to €500 million for a two-year period,' Professor Bachem explained. 'This is definitely too much for one country. And there is a need to consider this as one European research infrastructure with the support of all European countries.'

However, before that numerous challenges will have to be overcome, the most difficult one being legal rather than technical, said Professor Bachem: 'If we really want to build a European infrastructure, we have to decide which model best suits this infrastructure. Should it be a model like CERN [the European Organisation for Nuclear Research], should it be a model like ESA, the European Space Agency or any other European organisation? It is not easy to determine a suitable legal entity, which fits all the restrictions we have in the different countries.'

Technical challenges will be twofold: On the one hand, the researchers on the PRACE project will have to develop a suitable petaflop hardware system working with 100 000 processors. On the other hand, there is the software: 'We are not yet used to exploiting that many processors. So, we have to redesign our algorithms, maybe we also have to redesign our theories in order to tackle problems with that many processors.' Still, Professor Bachem is confident that the petaflop target will be met within the timeframe set.

But in as rapidly an evolving field as computers, this is obviously not the end of the rope. So, where will supercomputing be in 20 years? 'It's very hard to look that far into the future,' Professor Bachem stated, 'but I think in 20, 25 years we will be far beyond the exaflop, that is one step behind the petaflop. We may already have the first prototype of the quantum computer. If we have the same exponential growth that we have experienced in the last 20 years, there will be incredibly fast machines and with an incredible amount of space needed and these machines will look much different from the ones we have today.'