PRACE: Supercomputing Research Infrastructure for Europe

From AnnualReport

Jump to: navigation, search


Background

The Partnership for Advanced Computing in Europe (PRACE) is a pan-European Research Infrastructure for High Performance Computing (HPC) and forms the top level of the European HPC ecosystem. The infrastructure consists of several Tier-0 supercomputers (including one at BSC-CNS) distributed across the continent, providing outstanding computing services to enable world-class research on world-class systems. The systems are installed at centres in France, Germany, Italy and Spain. Through PRACE, scientists and technologists from Europe and around the world are provided access to world-class supercomputers with capabilities comparable to those available in the USA and Japan. These leadership class systems will help the continent’s scientists and engineers to remain internationally competitive.


BSC-CNS played a key role in the creation of the PRACE Research Infrastructure at all levels, with Sergi Girona serving as Chairman of the Board of Directors, significant technical contributions, and leadership of the organisational design. This included the selection of the best legal form, design of the governance structure, funding and usage models, and the peer review process, resulting in the PRACE statutes and initial agreement which were signed in May 2010. From that date, PRACE has operated as a Belgium-based legal entity (AISBL-Association Internationale Sans But Lucratif) in parallel to the supporting European projects.

BSC-CNS and PRACE

In May 2014, BSC-CNS hosted the first of the series of PRACEdays, the annual Scientific and Industrial conference of PRACE. The conference was addressed to scientists, industrial users, policy makers and HPC managers, for a better and common understanding of the needs, requirements and benefits of HPC for research, innovation and society.

BSC-CNS continues to operate as one of the six selected PRACE Advanced Training Centres (PATC). Each of the sites coordinates and executes training and education activities enabling the European research community to utilise the computational infrastructure available through PRACE. BSC-CNS is amongst the best performing PATC centers and in the 2013-14 academic year developed new courses based on tools and programming models developed at BSC-CNS.

Following completion of the PRACE 1st Implementation Phase project, another two European funded projects supporting implementation of the PRACE Research Infrastructure were run in parallel during 2014, all with a significant level of involvement of BSC-CNS. In the 2nd Implementation Project, BSC-CNS evaluated the latest hardware components installed in the previously selected prototypes, extended the community code enabling support, and continued research on best practice for HPC system commissioning and novel programming techniques. Finally, within the 3rd Implementation Project, BSC-CNS continued many of the PRACE 1st implementation phase PRACE activities including research on organisational aspects and the development of an impact assessment framework.

Within the framework of the PRACE Implementation Phase projects, BSC-CNS first deployed a 16-node cluster of hybrid ARM + CUDA GPU compute nodes to serve as a software development vehicle, and then deployed the first large-scale cluster using ARM processors driving the high-end NVIDIA Tesla accelerator. BSC-CNS also led and coordinated activities in four key research areas: auto-tuned and automatic techniques to be applied in parallel programming models, scalable numerical algorithms, performance tools, and file systems. BSC-CNS also significantly contributed to Community codes work in the fields of engineering and materials science with the most outstanding improvements achieved for Alya and SIESTA codes. The work for Alya consisted of parallelizing the partitioning of input meshes with PARMETIS. For SIESTA a solver was developed that reduces computational complexity without loss of accuracy. This was successfully tested on systems with tens of thousands of atoms, making new classes of physical problems accessible to first principle calculations.

Spanish participation in PRACE

Spanish scientists were very successful in participating in important research projects that were awarded with computing time on the various PRACE nodes.

Some of the projects with Spanish participation are:

  • Direct Numerical Simulation of Equilibrium Adverse Pressure Gradient Turbulent Boundary Layers, Prof. Javier Jiménez Sendín (UPM), 17 million hours on SuperMUC@LRZ@GCS
  • Large-scale radiation damage cascades from first principles, Prof. Emilio Artacho (CIC nanoGUNE), 36 million hours on Hermit@HLRS@GCS
  • INFLUM2 - Effects of selective mutations on the ligand binding and unbinding to the M2 proton channel of influenza virus, Prof F. Javier Luque (UB), 22 million hours on MareNostrum@BSC
  • NEMERTE - Numerical Experiment on the Mediterranean model response to Enhanced Resolution and TidE, with the participation of IMEDEA, 17 million hours on Fermi@CINECA
  • Identification of somatic variations in PanCancer genomes using SMUFIN, a reference-free approach. Prof. David Torrents, 6 million hours on MareNostrum@BSC
  • Protein-DNA binding allostery, Prof. Modesto Orozco (IRB), 29 Million hours on MareNostrum@BSC
  • EXCOMM -- EXtreme scale domain decomposition solvers for COMputational Mechanics, Prof. Santiago Badia (CIMNE-UPC), 10 million hours on Fermi@CINECA
Personal tools
Home