Fujitsu looks to maximise the performance of next-generation supercomputers

18 Nov 2010 | News
As supercomputing moves into the petascale, Fujitsu has launched a global collaboration to develop a mathematical library to support the next generation of supercomputers.

Fujitsu Limited and Fujitsu Laboratories of Europe Ltd have announced the launch of the Open Petascale Libraries (OPL) project, a global collaboration to develop a mathematical library that will serve as a development platform for applications running on petascale-class supercomputers.

The project, which will initially involve ten partners, including a number of universities and research institutions in Europe, will make the code developed publicly available in open-source form, for use by the computational science community as a whole.

In addition, the output from the OPL project will be applied to help accelerate the application development for Next-Generation Supercomputer known as the K computer, being developed in Japan, which is scheduled to begin operation 2012.

As a result, OPL is expected to make an important contribution to a range of fields, such as the life sciences, development of new materials and sources of energy, disaster prevention and mitigation, manufacturing technologies and basic research into the origins of matter and the universe.

Kimihiko Hirao, Director of Japan’s RIKEN Advanced Institute for Computational Science, said, “Supercomputing today is an invaluable foundation for advancing science and technology, and the scientific and technological achievements and knowledge gained through supercomputing will benefit humanity on many fronts. International collaboration is also increasingly important. This project follows this direction, and we aim to participate actively and produce meaningful results.”

Open Petascale Libraries Project: Aims and Objectives

The aim of the OPL project is to develop a mathematical library that will play an important role in each of the representative application areas for petascale supercomputers. Target systems for the library are the Next-Generation Supercomputer and x86 HPC clusters, which are standard systems used as supercomputers. The library’s parallelisation will adopt a hybrid parallel programming model, which is effective for today’s multi-core supercomputers. Using the code generated through this project, it will be possible for application developers to maximise the performance of petascale supercomputers.

Fujitsu will provide organisations participating in the project with technical information and a development environment. By making the code developed as part of the project available in open-source form, it is expected to be widely employed in the broad range of fields in which petascale supercomputers are utilised.

The ten organisations taking part include the Society of Scientific Systems in Japan, the Australian National University, Imperial College London, the Innovative Computing Laboratory at The University of Tennessee, The Numerical Algorithms Group in Oxford, UK, e-Research Centre, UK, The Science and Technology Facilities Council, UK, University College London, Fujitsu Limited, and Fujitsu Laboratories of Europe Ltd.

Petascale supercomputers, like the Next-Generation Supercomputer, are capable of quickly performing large-scale and advanced computations that are difficult to solve using normal computers. In order to maximise the performance of petascale supercomputers - which perform massive-scale parallel computations by linking tens of thousands of processors, each featuring many computational cores - it is necessary to develop applications that can efficiently coordinate hundreds of thousands of computational cores and smoothly perform these parallel computations.

It has become a significant challenge for computational scientists to develop these applications. A powerful approach to overcoming this complexity is to develop a common mathematical library that can be employed by applications in each area. Developing such a mathematical library for petascale supercomputer applications requires far deeper knowledge of computer architecture and applications compared to the mathematical libraries used in existing supercomputer applications, underlining the significance of the OPL project’s collaborative approach.

Never miss an update from Science|Business:   Newsletter sign-up