Europe gets its HPC act together
At ISC’13, Tom Wilkie heard some European success stories in research using high-performance computing and learned of ambitious plans for a European HPC industry. But he also heard some criticisms too – are there dark clouds on Europe’s HPC horizon?
In a neat bit of symmetry, European cooperation bracketed this year's International Supercomputing Conference and Exhibition (ISC'13) held from 16 to 20 June in Leipzig, Germany. Academic collaboration, in the form of the Partnership for Advanced Computing in Europe (PRACE), was the subject of a mini-conference on the opening day; while at a breakfast meeting on the last day Europe's HPC industry outlined its Strategic Research Agenda to develop European industrial capacity in actually making supercomputers, the software that operates them, and the application software that makes them useful.
The PRACE seminar heard some outstanding success stories of European research – directly relevant to industry, even though carried out in academic environments – that could have been achieved only through the infrastructure provided by the PRACE network. What could be more pan-European than an Italian meteorologist, working at the UK's Reading University, and carrying out climate-model calculations on a German supercomputer located in Stuttgart? According to Per Luigi Vidale, professor of Climate System Science at Reading, it would have taken about ten years, using the computational facilities available in the UK, to develop his models to the point where he can now consider 1km resolution. Only at this scale is it possible to model the effect of clouds properly, and of course, clouds are vital to a fuller understanding of weather and of climate change.
He pointed out that the predictions of such global models are not just of academic interest but are vital to industry. He cited the example of heavy rainfall in the Philippines in 2005 that led to the flooding of a factory making hard disks. This had global economic repercussions as many computer manufacturers were unable to source drives from elsewhere and global supply chains were disrupted. But the meteorological event was a complex one, where an El Nino far to the east on the other side of the Pacific combined with a more local typhoon to produce the exceptional rainfall in the storm that hit the Philippines. Weather and climate models need to be global in scope if they are to forecast such events with any degree of accuracy.
He warned that global models are predicting that there will be fewer tropical storms in the future, as a consequence of climate change, ‘but they will be more intense’. A further exceptional event that illustrated the interconnectedness of the Earth’s weather systems was the European heat-wave of 2003. Temperatures rose to such levels that thousands died. According to the models, this was indeed an exceptional event and is not likely to become frequent. It was due to a combination of factors, including low soil moisture levels that year resulting from lighter rainfall in the previous year, but once again events as far away as the Pacific Ocean had contributed to this exceptional behaviour of the weather in Europe.
Business and industry, as well as Governments, need access to the results of such global models quickly and easily if they are to prepare for, or hedge against, the risk of future supply chain disruption or civil contingencies, he said. But as the models peer down to finer and finer scales, the computers need to be faster and more powerful if they are to produce results within the timeframe that industry requires.
Europe's consumption of high-performance computing (HPC) far outstrips its commercial and industrial production of HPC technology, hardware, and software, according to Marcin Ostasz from the Barcelona Supercomputing Centre. The share that European vendors have of the global HPC systems market is less than five per cent. To try to rectify the imbalance, many of the leading companies in Europe's high-performance computing industry have come together to form a European Technology Platform for HPC (ETP4HPC). It published its Strategic Research Agenda (SRA) in May 2013, just before the ISC'13 conference, and held a breakfast meeting on the last day of the conference to outline its strategy to build up Europe’s capacity to provide a complete HPC supply chain.
The group is proposing that Europe should commit to a research and development programme worth 150 million Euros a year, as part of the European Commission's Horizon 2020 R&D strategy. Half the funding should come from the Commission and half from other sources, and the group appears confident that the Commission will accept its proposals and that the first ‘public private partnership’ will get underway later this year.
Jean-Francois Lavignon of Bull, who is chairman of ETP4HPC, told the meeting that the aim was to develop European leadership in HPC. To achieve this, required three strands to the programme, he said: the first was to develop HPC technologies in Europe; the second to have a world-class European HPC infrastructure; and the third was to develop HPC applications and encourage their widespread use. Key to successfully achieving these aims, he continued, would be coordination of efforts across Europe; education and training of personnel both in the underlying technologies but also in parallel programming and making use of the technologies; and finally a ‘focus on SMEs’ – encouraging small and medium enterprises to adopt high-performance computing techniques in their businesses. (As an aside, throughout the whole of ISC’13 Intel had hung a banner across the entrance to the conference halls with the message: 'To compete, you must compute'. Although that exact phrasing was not used by the ETP4HPC, their underlying message was very similar.)
Part of the action plan, as M. Lavignon outlined it, would be the creation of a European Centre of Excellence that would encourage application development. Although it is not yet completely defined how this will be structured, he said, the Centre is clearly intended to be a large undertaking, potentially with more than 50 professional staff, and an annual budget by itself of more than five million Euros. As a European project, it would derive some financial support from the European Commission. However, it is envisaged that the European Centre of Excellence would be complemented by smaller ‘Centres of Competence’ organised on a regional or country-wide level. Whether these would receive EC funding, or be reliant on national Government and regional funds, was not clear.
The group has identified some 140 milestones for the Strategic Research Agenda, and it is recommending that the programme should be funded in two phases, The first phase would last three years and would deliver results useful for the next-generation of ‘extreme scale’ computer systems as well as for HPC solutions that address a wide market. The results of the first phase would feed into the design of the second tranche of research projects that would aim to deliver technologies for Exascale as well as HPC systems that are more energy-efficient.
The academic programme, PRACE, has been a pathfinder in organising cooperation in the use of HPC resources in Europe. But PRACE has also revealed an imbalance in the HPC infrastructure within Europe itself. The big machines are in Germany and there are few providers of capacity elsewhere. While the scientific achievements are unquestioned, some murmurs could be heard in the corridors of ISC’13, both about this geographical imbalance but also about the growth of bureaucracy. PRACE has recently had to formalise its structure – incorporating itself as an international not-for-profit association (Association internationale sans but lucratif or AISBL) in Brussels, Belgium, and the inevitable administration involved in managing such a complex legal entity has led to sotto voce worries about bureaucracy stifling creativity.
There is always a tension between scientists’ drive to tackle the next big research question and the bureaucrats trying to find the funds to pay for it. But perhaps more worrying for the future health of European HPC, were the questions being quietly asked, outside of the formal conference sessions, about the geographical imbalance within PRACE. For how much longer, it was asked, would German politicians be content to allow other Europeans access to HPC infrastructure paid for by the German taxpayer, if they could not see an equivalent benefit for German researchers elsewhere in Europe?
The European Commission appears to recognise the importance of high-performance computing to Europe – the results of computational models are vital to science, engineering, and industry while the underpinning technology is something that Europe needs to develop if it is to remain an advanced industrial region. If member Governments share that vision, then they will find ways to answer these criticisms, but if in a time of austerity they pull back, Europe may find itself neither computing nor competing.