2024 InPEx Workshop

Work group description

The goal of the AI for Scientific Computing group is to advance the field of Scientific Artificial Intelligence by fostering an environment of collaboration and innovation among core AI teams at international level. The group aims to share practical use-cases and benchmarks, including datasets and metrics, that underline the critical aspects of AI in scientific domains.

The group will focus on key strategies for advancing AI in scientific fields. An emphasis will be put in the formation of collaborative teams to develop trustworthy, scientifically validated AI models with clear, reproducible results. The approach of the group includes advocating for common practices and frameworks essential for creating foundation models, alongside defining and sharing use-cases with access to crucial datasets and metrics. The initiatives also extend to creating challenges to foster new community involvement and developing AI models for handling complex scientific data. Efforts will also be done to integrate AI with HPC to improve scalability and extend the scope of AI solutions to include both large-scale systems and smaller, hybrid models.

Work group description

The upcoming conference session is designed to enhance the exchange of innovative software components and development methods, particularly within computational mathematics and software engineering. It aims to promote international collaboration on software development kits (SDKs) geared towards application-driven purposes, emphasizing collaborative continuous integration and performance analysis. The session will also focus on efficient discretization of Partial Differential Equations (PDEs) and Block-structured Adaptive Mesh Refinement (AMR) at the exascale level.

Our goals include optimizing and expanding the use of existing software tools found in ECP SDKs like CEED and AMReX. We plan to form teams and communication networks to enhance software interoperability and introduce new frameworks. The session will feature the development of standardized proxy-apps and mini-apps within these SDKs to facilitate shared methodologies for performance assessment and the dissemination of insights. Additionally, it will underscore the significance of refining performance portable programming models such as Kokkos to meet application-specific needs and support sustainable software initiatives like the High-performance Software Foundation (HPSF). Ultimately, we seek to advance exascale application methodologies through the use of superior software components and cutting-edge programming models.

Work group description

The “Digital Continuum and Data Management” subgroup held its first reflection session in the Reims meeting, 2023. This subgroup focuses on the continuum of technologies and practices related to managing data from collection to processing and storage, including edge computing and cyber-physical systems.

Currently, the Digital Continuum is primarily being driven by large cloud providers, while high-performance computing (HPC) centers are competing with these providers to offer alternative solutions for managing and processing large datasets. One of the challenges facing HPC centers is demonstrating their value proposition compared to cloud providers, especially when it comes to real-time data processing requirements.

Another challenge for the Digital Continuum is that it is a multi-tenant environment where collected data is used for multiple purposes and computing infrastructure is shared. As such, building trust in this continuum of entities is crucial and the related issue of cybersecurity becomes central.

To address these challenges, the subgroup Digital Continuum proposes for the June 2024 INPEX meeting to focus on building a continuum of trusted entities as a proof-of-concept, using multiple infrastructures. The key to achieving this goal is finding good candidate applications that can demonstrate the value and capabilities of the Digital Continuum.

To this end, the subgroup proposes to discuss potential approaches using real use-cases during the meeting.

Work group description

Exascale and Post-Exascale applications are becoming increasingly difficult to build, deploy and maintain under the double pressure of the growing machine complexity and the application needs to combine multiple compute and data processing paradigms (HPC+HPDA+AI). To address these challenges, our community needs to foster  HPC dev-ops methodologies and tools to enhance productivity and enforce  interoperability, functional and performance portability as well as reproducibility. Topics of disucssion include (but not limited to):

• Package Managers and Containers – Empowering users beyond Modules
• Shared Repos, Build Farms and Caches for Modules/Packages/Containers – Frictionless Access – Security
• CI/CD at Exascale – Testing – Storage and Compute Costs
• Automation – HPC dev-ops – Everything-as-Code – Everything-as-a-Service
• New software stacks (AI)
• Legacy code – Fortran compilers – LLVM
• AI-assistants for development, debugging, documentation,…
• Catalogs and Metadata
• Debugging at Exascale
• Componentisation – Interoperability – Standardization – Common software stacks
• Software Sustainability