Activities

EA-MHPC White Paper

Read online below or download PDF>>

As engines for cross-sectoral digital transformations in many scientific, economic and social fields, Exascale and post Exascale supercomputers will be key to advancing automotive, aerospace, chemical, healthcare and energy industries and to strengthening specific European competences in materials science, molecular biology, personalised medicine, neuroscience, AI as well as climate and earth system sciences, in particular. Critical technological building blocks and integration paths are the critical ingredients to realize the objectives of the European High Performance Computing Joint Undertaking (EuroHPC-JU), i.e., the development of a hyper-connected set of Exascale and post Exascale supercomputers for high-end simulation and processing of big data, which can be spearheads and part of a wider vivid HPC ecosystem.

This document describes the high-level strategy — including concrete key elements and efforts — needed to realize a vivid ecosystem of European supercomputers, the prospects for a lasting European HPC technology, and the role of pilot platforms in this context. Key assets are the successful Sequana architecture and environment, co-developed by ATOS and CEA, and the proven modular middleware ParaStation Modulo, co-developed by PARTEC and FZJ. A third element is constituted by hardware components developed in Europe, with the EPI processors and the BXI network as major examples. The EPI project was initiated by ATOS, BSC, CEA, and FZJ with many additional partners, including SiPEARL, which will industrialize the EPI general purpose processor. The BXI network initially co-designed by ATOS and CEA is also part of multiple EuroHPC R&I projects. The Modular Supercomputing Architecture (MSA), a truly European approach to heterogeneous computing, combines all these elements putting them at the service of a wide variety of application fields and satisfying their diverse requirements.

SEA projects

The EuroHPC projects DEEP-SEA, IO-SEA and RED-SEA are examples of EA-MHPC founders’ commitment and actions to develop complementary European technologies for heterogeneous exascale supercomputing architectures.

The SEA projects aim to drive further the vision shared by the Jülich Supercomputing Centre (JSC), CEA (the French Alternative Energies and Atomic Energy Commission), Atos and ParTec, of the modular supercomputer architecture. The four core partners have gathered around them the best European skills, whether academic or industrial, to strengthen the European technologies that will equip the future European Exascale computers.

The SEA projects build upon the results of past and currently running European projects. They will particularly align their efforts with the hardware-focused projects such as the European Processor Initiative, with the ambition to contribute to a complete and sustainable European High Performance Computing (HPC) ecosystem for the exascale architecture.

The three SEA have been accepted by the EuroHPC JU under its Work Programme 2019, and started on 1^st April 2021, for a duration of three years each.

DEEP-SEA is the fourth project in the DEEP project series and has set as its goal to improve the manageability and programmability of highly heterogeneous system architectures. In particular, DEEP-SEA will make it easier to program and run applications on the Modular Supercomputer Architecture, which was introduced by the DEEP-EST project and is already implemented in the JUWELS system at the Forschungszentrum Jülich and in the EuroHPC MeluXina system in Luxembourg.

IO-SEA will cover the Input/Output and data management aspects that will help tackle the increased pressure on storage systems in the context of exascale systems. It aims to design and develop a novel management and storage platform based on object stores, allowing for hierarchical storage management (HSM) and on-demand provisioning of storage services. It leverages the results from the Sage, Sage 2 and Maestro projects.

RED-SEA will look into the network interconnect – a potential bottleneck at the Exascale. It aims to extend and optimize the production-proven European interconnect BXI (BullSequana eXascale Interconnect) to make it capable to power the future European Exascale systems. As HPC systems increasingly need to interact securely with the outside world, including public clouds, edge servers or third party HPC systems, RED-SEA will also add a seamless interface with Ethernet. The project relies on technology developed within previous projects such as ExaNest and Mont-Blanc.