Insight into European Supercomputing - The DEISA Extreme Computing Initiative
By Wolfgang Gentzsch, Alison Kennedy, and Hermann Lederer, DEISA Project, Germany
DEISA is a consortium of leading national Supercomputing Centres in Europe which has designed, deployed and operates the Distributed European Infrastructure for Supercomputing Applications at a continental scale. Supercomputing services include operation, technology, and application support. The aggregate peak performance of the complex and heterogeneous supercomputing environment exceeds two PetaFlop/s. Supercomputing is a key technology for scientific research with high societal impact (see this issue’s article on Economic Competitiveness in the 21st Century by Cynthia McIntyre). DEISA reflects the collaborative European approach of pooling national high performance computer resources and making them available for the most challenging computational projects in a unified infrastructure together with well structured services. The DEISA initiative was formed in early 2002, while the DEISA EU FP6 project started in 2004 and was continued in 2008 through the EU FP7 DEISA-2 project.
Flagship service for European scientists is DECI, the DEISA Extreme Computing Initiative, to enable “grand challenge” applications in all areas of science and technology. As of 2005, challenging computational science projects have been conducted on a regular basis. Over 180 different research institutes and universities from 25 European countries with more than 360 researchers have participated in DECI to date, including collaborators from North and South America, Asia and Australia. To further strengthen European multi-national scientific collaborations, support was introduced for science communities as a whole in addition to the single project oriented support through the DECI. DECI and science communities are supported by applications experts from leading European HPC centres.
Thus DEISA enables European computational scientists to obtain access to the most powerful national computing resources in Europe regardless of their country of origin or work. Also smaller countries are enabled to participate in cutting-edge research through internationally competitive computational science.
Insight into the differences between usage of national and European resources and facilities gained by DECI also facilitates the understanding of requirements of users of future European leadership-class supercomputers installed by PRACE. Through an annual call, a number of high-performance computing projects are selected by peer-review on the basis of innovation and scientific excellence. Successful projects are given access to the exceptional resources in the DEISA infrastructure and are offered applications support to enable them to use it productively. The number of scientific proposals received has grown over the years, from 51 at DECI-1 to 122 at DECI-6, with particularly rapid growth over the past two years.
The DECI-5 call in 2009 attracted 75 proposals, requesting 220 million compute-hours together with significant application support. 69 million compute-hours were allocated to 50 projects. The most recent call (DECI-6) attracted 122 proposals and has requested more than half a billion compute-hours.
We can show how both the median (the middle of a distribution) and the average amount of CPU requested by projects has grown year on year. The growing divergence between the average and the median CPU requested reflects the increasing number of larger, collaborative projects applying to DEISA for computational and applications enabling resources. For example, particle and plasma physics projects (where there is a long tradition of cooperation at the European and international level) generally requested many more resources on average in DECI-5 than other applications areas. The materials science community, however, is much more disparate and submitted a much larger number of smaller projects.
Overview of DECI-5 Projects
The 50 DECI projects selected for DECI-5 were from 6 scientific disciplines – Astro Sciences, Bio Sciences, Earth Sciences, Engineering, Materials Science and Plasma and Particle Physics. 70% of the investigators came from countries with a DEISA partner site, 21% from other European countries, and 9% from outside of Europe (USA, China, Israel, and Japan). This suggests that successful project proposals were generally based on collaborative European science.
The number of investigators in DECI-5 projects varied from one to ten per project, with the median number being three. During the five years of DECI operation, the average number of investigators per project has risen steadily from 1.0 to 3.0.
The vast majority of successful projects (66%) involved scientists from more than one country, with more than a quarter (26%) involving scientists from three or more countries. This indicates that DECI is supporting pan-European scientific collaboration, and that the DEISA infrastructure is attractive to European researchers. Of the scientific investigators who have used the DEISA infrastructure via DECI so far, 77% have been involved in one project to date with a further 23% involved in two or more projects. This shows that DEISA is successful in reaching out to new groups and collaborations. It also suggests that DEISA resources are recognised as being complementary to national facilities, with applications only being made to DEISA when the circumstances warrant it.
A total of 66 different computer codes were specified for the 50 projects, indicating the wide variety of science
being undertaken via DECI. The need for close partnership with users to achieve good performance via applications enabling and code tuning is reinforced. There is obviously a strong demand for the sort of complementary computing support offered by DECI which crosses scientific domains.
