HPC Globalization Fuels ISC Growth
This year’s International Supercomputing Conference (ISC’09) in Hamburg, Germany, promises to offer some respite from the gloomy news affecting the HPC community in these tough economic times. According to ISC’09 organizers, overall registration is up 20 percent over the 2008 conference. With upwards of 1,500 people expected to attend and a record 116 exhibitors having signed up to display their HPC wares, the event is on track to be the largest in its 24-year history.
Given the severe downturn in the global economy, and the recent rash of HPC company shutdowns and acquisitions, this is good news indeed. And not just for the broader HPC community. After moving to the conference to Hamburg for 2009, the ISC organizers are undoubtedly breathing a sigh of relief that their choice of a bigger venue has been vindicated.
Some long-time ISC attendees may wish for the intimate setting in Mannheim, where the conference began in 1986, and later in Heidelberg. During the early years, conference chief Hans Meuer personally hosted dinner for 60 or 70 people at local inns, where they watched European Cup soccer and partook of authentic German beer and food. At that time, it was more of a regional conference, much more focused on supercomputing research and academia, and distinctly German in flavor.
Those days are gone. With ISC’s increasing footprint, in some ways the European conference is starting to resemble the truly super-sized SC event in the US (which still manages, however, to attract a much larger crowd — over 300 exhibitors and more than 10,000 attendees). But like SC, ISC is now filled with vendor presentations and exhibits, Birds of a Feather (BoF) get-togethers, poster sessions, and breakout meetings, all of which are running concurrently with the main conference sessions.
What’s behind the ISC growth? The same forces driving high performance computing expansion in general — globalization and the “democratization” of HPC. Together they are flattening the HPC landscape worldwide, and enabling European and Asian countries to catch up to what has been American dominance in HPC. India and China are just beginning to invest heavily in supercomputing, but their big growth spurt is still to come. In Western Europe and Japan, the use of high performance computing for both government and commercial purposes has been widespread for some time, but now they’re closing the gap at the high end. Japan is backing a plan to get to exascale computing by 2020 and Germany is investing nearly $1 billion in next-generation hardware.
In May, Europe’s first petaflop supercomputer, the Blue Gene/P JUGENE system, was inaugurated at the Jülich Supercomputing Center, in Germany, along with JUROPA (207 teraflops) and HPC-FF (101 teraflops). The latter two machines are closely coupled and are slated to be combined for certain types of fusion research work. And in Switzerland early this month, the Swiss National Supercomputing Centre upgraded its Cray “Monte Rosa” supercomputer, increasing its peak performance to 140 teraflops. Also this month, the UK’s HECToR system is a getting a makeover, pushing its peak output to 200 teraflops.
On the software side of HPC, there is also a growing degree of international parity. As pointed out by the recent World Technology Evaluation Center (WTEC) study [PDF] that assessed global research and development in simulation-based engineering and science (SBE&S), many countries are catching up and even surpassing the US in certain areas. The report points to Europe in particular as an area where SBE&S software research is getting a lot of attention:
Aggressive, well-funded initiatives in the European Union may undermine US leadership in the development of computer architectures and applied algorithms. Examples of these initiatives include the Partnership for Advanced Computing in Europe (PRACE) which is a coalition of 15 countries and led by Germany and France, and based on the ESFRI Roadmap (ESFRI 2006); TALOS – Industry-government alliance to accelerate HPC solutions for large-scale computing systems in Europe; and DEISA – Consortium of 11 leading European Union national supercomputing centers to form the equivalent of the US TeraGrid. There is also some flux, with some alliances dissolving and new consortia being formed. Already, the European Union leads the United States in theoretical algorithm development, and has for some time; these new initiatives may further widen that lead and create new imbalances.
Globalization also means that many HPC providers now have international footprints. The two biggest HPC system vendors — IBM and HP — are completely globalized. While both firms remain based in the US, they each maintain facilities in Europe, including research labs. According to Horst Simon, associate laboratory director for computing sciences at Berkeley Lab, most Europeans now look upon IBM and HP as local vendors. He believes the advantage the US had with American vendors has gradually diminished. “Ten or fifteen years ago, we might have said that IBM was an American company and would have dealt with the US labs first,” explains Simon. “Now the world is flat and a customer is a customer.”
The US still has a few unique advantages, though. Since American policy-making is executed under the federal government, its national labs are much better organized in comparison to the more hodge-podge makeup of the research centers in the European Union. To its credit, PRACE is attempting to correct this by coalescing the resources of 15 EU countries in order to build a competitive HPC infrastructure in Europe. Its initial goal is to bring five petascale systems online, beginning this year.
Whether the recession slows down the global advance of HPC remains to be seen. If it does, it will only be a temporary speed bump. As ISC shows, international interest in high performance computing is growing, and given the problems of the 21st century that can be solved with the help of HPC, it is likely to continue to do so.