First NumPEx/Exa-DI workshop on Efficient Discretisation for PDE@Exascale

[prudhomm]

Workshop Efficient Discretisation for PDE@Exascale

The first workshop of the Exa-DI project, a project of the NumPEx programme, "Efficient Discretisation for PDE@Exascale," took place on November 7 and 8, 2023. It was held at the Amphithéâtre J. Talairach (Neurospin) at CEA Saclay in Gif-sur-Yvette.

Description

This workshop featured sessions on various aspects of exascale computing and its applications, including participant introductions, technical presentations, discussions, and roundtable sessions. The event brought together experts to discuss advancements and future directions in discretizing partial differential equations (PDEs) efficiently for exascale applications.

The workshop "Efficient Discretisation for PDE@Exascale" featured a diverse group of participants presenting Application Demonstrators from various research and industry sectors. Key contributors included:

1. Jean-Pierre Vilotte from CNRS, who provided the introductory context for the workshop.
2. Eric Savin from ONERA, discussing exascale performance evaluation for aircraft prototypes.
3. Henri Calandra from TotalEnergies, focusing on exascale multiphysics simulators for CO2 sequestration.
4. Christian Trott from SNL, presenting on Kokkos.
5. Julien Vanharen & Loic Marechal from INRIA, addressing exascale simulations for turbomachinery.
6. Nathanaël Schaeffer & Hugo Frezat from CNRS - Université Grenoble Alpes & Paris Cité, exploring machine learning applications in Earth dynamo simulations.
7. Vincent Chabannes & Christophe Prud’homme from Université de Strasbourg, Cemosis and CoE Hidalgo2 discussing dynamic energy simulation for urban buildings.
8. Olivier Jamond from CEA, presenting the MANTA project.
9. Soleiman Yousef from IFP Energies nouvelles, on performance issues in geoscience applications.
10. Virginie GrandGirard from CEA, discussing the GYSELA code for plasma turbulence simulations.

Key sessions included:

1. Introduction and Context: Presented by Jean-Pierre Vilotte from CNRS, setting the stage for the workshop's main theme.

2. Participant Self-Introduction: Allowing attendees to introduce themselves and their interests.

3. Various Technical Sessions: These sessions featured talks on topics like exascale performance evaluation for aircraft prototypes, multiphysics simulators for CO2 sequestration, and advancements in exascale simulations for different applications like turbomachinery and dynamic energy simulation for urban buildings.

4. Discussions and RoundTables: These sessions provided opportunities for attendees to engage in discussions and share insights on the presented topics.

NumPEx

The PEPR NumPEx, standing for "Programme et Équipement Prioritaire de Recherche Numérique pour Exascale" (Priority Research Programs and Equipment for Exascale in Digital and Computing), is a significant French initiative focused on advancing the fields of large-scale and exascale computing. This exploratory research program is spearheaded by prominent French research institutions: CNRS, CEA, and Inria.

The primary goal of NumPEx is to design and develop the necessary software components or building blocks that will be integral to future exascale machines. Exascale computing represents the next significant leap in computing power, capable of performing a billion billion (i.e., a quintillion) calculations per second. This immense computational capability is crucial for tackling complex scientific, industrial, and technological problems.

NumPEx plays a pivotal role in equipping these future exascale machines, ensuring they are ready for a wide array of major application domains. By doing so, it aims to fully exploit the capabilities of these advanced machines to benefit scientific research and various industrial applications. Additionally, the program is an integral part of the broader European Exascale roadmap, highlighting its significance in the context of continental scientific and technological progress.

Numpex Projects

The NumPEx program encompasses five key projects, each focusing on different aspects of Exascale computing:

1. Exa-MA: Methods and Algorithms for Exascale

   • This project addresses the need for advanced methods and algorithms suitable for exascale computing. It aims to develop predictive models and analyze massive data quantities using AI, thereby creating new scientific opportunities. The project focuses on avoiding communication bottlenecks, adapting to heterogeneous hardware, and improving scalability and self-parametrization of numerical methods.

2. Exa-SofT: HPC Software and Tools

   • Exa-SofT is dedicated to evolving the HPC software stack to better support the heterogeneous and scalable architectures of exascale supercomputers. The project addresses challenges like productivity, performance portability, heterogeneity, scalability, resilience, and energy efficiency, with the aim of consolidating the exascale software ecosystem through multidisciplinary research collaborations.

3. Exa-DoST: Data-oriented Software and Tools for the Exascale

   • The Exa-DoST project focuses on the scalability of techniques used for data storage, transfer, processing, and analytics in the context of exascale computing. It aims to develop solutions for scalable storage and I/O, in situ processing, and smart analytics, addressing the data challenges and ensuring operational solutions for the HPC software stack.

4. Exa-AToW: Architectures and Tools for Large-Scale Workflows

   • This project aims to provide efficient management solutions for large-scale workflows that involve HPDA, AI, and HPC tasks. Exa-AToW focuses on end-to-end solutions considering workflows, data logistics, resource federation governance, cybersecurity, and sustainability. It develops a data-driven, service-oriented architecture integrating technologies with governance and cybersecurity, considering global optimization for performance, energy, and sustainability.

5. Exa-DI: Development and Integration

   • Exa-DI brings together research, software, and applications in an iterative and incremental process supported by a computational and data team (CDT). It aims to accelerate the delivery of a co-designed and co-developed science-driven software stack. Exa-DI focuses on the "last mile" integration of logical collections within the NumPEx software stack and supports this with software community policies, common exascale software packaging technologies, and a software integration hub. Its Agile methodology drives the process to improve exascale application capability and performance portability.

These projects collectively aim to advance the capabilities and applications of exascale computing, addressing the various challenges and opportunities in this rapidly evolving field.