Participants in HOPSA include T-Platforms, Research Computing Center of MSU, Joint Supercomputer Center of RAS, Southern Federal University, Juelich Supercomputing Centre, Barcelona Supercomputing Center, German Research School for Simulation Sciences, Rogue Wave Software AB, and Technische Universitaet Dresden.
Whether analysing complex molecules, searching for new medically active substances, calculating the global climate or modelling astronomical events - computer simulations are becoming an indispensable tool in an increasing number of scientific fields. New more powerful supercomputers enable more realistic and more detailed simulations of complex global processes, whereas at the same time it is becoming more and more difficult for researchers to obtain an overview of all these processes and to identify sources of error or performance bottlenecks. Today the fastest supercomputers have tens or hundreds of thousands of processors working in parallel which, if possible, have to be utilized uniformly during the course of a simulation. In order to help users optimize performance more easily, Russian and European experts have established HOPSA, the first programme of its kind to consider all aspects of supercomputer performance analysis - ranging from running applications down to the hardware actually used.
The HOPSA project is being funded for two years as part of the EU's Seventh Framework Programme (FP7) and by the Russian Ministry of Education and Science in a special collaborative EURussia funding programme. The Russian research groups and developers are concentrating on tools for controlling and optimizing system performance while the European teams are contributing their expertise for developing and monitoring parallel applications in the project.
Efficiently solving large-scale important scientific simulations is getting more difficult every day because of the constant changes in supercomputing architecture and parallel programming and the ever growing complexity of interactions between hardware, system and application software components", stated Bernd Mohr, lead of the programme optimization team of Juelich Supercomputing Centre of Forschungszentrum Juelich, Germany's largest national multi-disciplinary laboratory. The HOPSA project, in which we are jointly involved with our Russian and European colleagues, will solve this problem by providing a unique holistic performance analysis and optimization work flow including all components involved in the process."
"Boosting computing systems' efficiency is a key goal throughout the global HPC community. Now, with leading Russian and EU experts combining their skills and resources to resolve this, I believe we can expect a significant breakthrough in this area", stated Vsevolod Opanasenko, CEO, T-Platforms.
"HOPSA-enabled solutions will dramatically improve existing supercomputers' performance along with providing new solutions to upgrade the technology base of the next generation of systems being designed now."
HOPSA is aimed at developing a unique unified infrastructure for holistic performance analysis including system-level and application-level monitoring. The Russian HOPSA collaborators will focus on the computing systems' hardware performance and settings, while their European partners will tackle optimization of parallel applications with the benefit of the combined data from system-level monitoring and application monitoring. HOPSA will deliver a set of integrated tools for those tasks, including well-established commercial products such as Vampir and ThreadSpotter as well as renowned Open Source solutions such as Extrae/Paraver, Dimemas, Scalasca, CUBE, and VampirTrace.
The latest results from the HOPSA Project are being showcased from 19 to 23 June at this year's International Supercomputing Conference (ISC) in Hamburg, the leading European supercomputing fair. Many of the project partners will be presenting their products there.