EESI projects to assess the European exascale challenges


25 Jun 2014 Leipzig - In the session "HPC in Europe" at ISC'14 in Leipzig, Stéphane Requena from GENCI talked about the exascale challenges of European academic and industrial applications. The goal is to build an international plan for developing the next generation of open source software for scientific HPC in the EESI projects 1 - which has already finished - and 2, both funded by the European Commission.

The EESI 2 project will define and assess the challenges for new technologies for exascale, Stéphane Requena stated. Software maturity and international training are important elements in this roadmap. The EESI 2 project is working together with PRACE to this purpose.

The EESI 2 project has a duration of 30 months and started on 1 September 2012.

Exascale applications include climate research; the Human Brain Project; industrial applications; and genomics.

Stéphane Requena insisted that we are not only talking exaflops here but also exabytes. Scientific research includes the generation of a lot of data. Extreme computing and Big Data are going hand in hand.

Some exascale industrial roadmaps are being set up. Boeing, for instance, is trying to reduce the wind-tunnel tests using HPC-enabled CFD runs instead for a better representation of aerodynamics, Stéphane Requena explained. This phenomenon turned into better design choices. All in all, digital aircraft design will require at least 1 Zetaflops.

Stéphane Requena cited a few challenges including meshers, solvers, resilience, couplers, optimization, frameworks, data and ISV licenses/scaling versus OpenSource.

Turning to the academic applications, Stéphane Requena expanded on climate research. We are heading towards coupled multiphysics climate models for the IPCC campaigns. The 5th IPCC campaign was hosted in France. The software challenges include the development of efficient couplers for multiscale and multiphysics; data management; and uncertainties quantification frameworks.

Stéphane Requena went on to sketch the global roadmap for future exercises in the area of astrophysics and cosmology. Initiatives such as ALMA, VLT, Euclid and also SKA require a dedicated HPC system of up to 1 Eflops. The software challenges in these projects also include the development of efficient couplers for multiscale and multiphysicis; data management; and the scalability, use of accelerators, resilience, load balancing versus smart runtime.

In the life sciences we are dealing with molecular and organ simulation. In this area, the software challenges include multiscale simulations; data management, in situ data analysis and visualization; and interactive supercomputing, Stéphane Requena explained.

Europe owns a big part of the applications used in the world, forn instance, up to 70% in chemistry. Some major applications breakthroughs recently have been performed in Europe. Stéphane Requena mentioned the massive allocations on the PRACE systems; the availability of large industrial HPC systems; and the extreme scaling of European Union applications.

In summary, he repeated that exascale is not only about exaflops but also about exabytes. Only a few heroes' applications are scalable to exascale so we need to support both capability and capacity simulations, as well as rethink and rewrite the applications and incorporate legacy applications for multi-scale and multi-physics simulations and for multi-disciplinary optimization.

The EESI 2 project recommends the use of ultra scalable algorithms and acknowledges the importance of resilience, Big Data, couplers, high productivity programming models, mini applications, software engineering methods for HPC, and uncertainty quantification verification and validation.

You can consult the EESI documents at http://www.eesi-project.eu

Leslie Versweyveld