SCA/HPCAsia 2026, Osaka, Japan

Full-day PDI tutorial

Decoupling I/O in HPC Codes with PDI: From File-based I/O to In-Situ Data Analytics

As HPC simulations scale toward exascale, I/O becomes a critical bottleneck due to the growing disparity between compute performance and storage bandwidth. Traditional post-hoc output models struggle with the volume and velocity of generated data. The Parallel Data Interface (PDI) offers a lightweight and flexible solution by decoupling I/O, filtering, and analysis logic from the simulation code. Through a declarative configuration system and a plugin-based architecture, PDI enables simulation developers to expose data buffers and trigger events without embedding I/O decisions directly into their application. PDI offers a simple API in C/C++, Fortran, and Python. This full-day tutorial introduces PDI and its ecosystem of plugins: sequential and parallel HDF5 for file output, user-code and pycall for in-process custom logic execution, DEISA for in-situ analytics, and Catalyst for live ParaView-based visualization.
Tutorial takeaways:
Through a combination of theoretical lectures and hands-on exercises, participants will learn how to instrument a simulation code with PDI, configure it declaratively via YAML, and drive complex I/O and visualization workflows without modifying the simulation itself. Attendees will leave with a practical understanding of how to adopt PDI in their own projects to create modular data workflows suited for HPC.

Additional hero image

Team

Benoît Martin, PhD . CEA/MdlS

Benoît Martin is a Research Scientist at Maison de la Simulation, a research laboratory affiliated with CEA, CNRS, Université Paris-Saclay, and Université Versailles Saint-Quentin. He earned his PhD in distributed systems from Sorbonne Université in Paris, under the supervision of Marc Shapiro. His academic background includes a BSc and an Engineering degree from École Supérieure de Génie Informatique (ESGI) in 2012 and 2015, respectively, as well as an MSc from Université du Québec à Chicoutimi (UQAC) in 2015. Before pursuing his PhD, he worked as an engineer in high-performance computing and the fintech industry, and as a research engineer at Inria with the Aramis team. His current research focuses on in-situ analytics in high-performance computing, particularly addressing challenges related to distributed systems and programming models. His interests include Transactional Memory Models, Consistency Models for Shared Memory, Conflict-Free Replicated Data Types (CRDTs), and the Actor Model.

Jacques Morice, PhD. CEA/MdlS

Jacques Morice is a Research Engineer at Maison de la Simulation, a collaborative laboratory involving CEA, CNRS, Université Paris-Saclay, and Université Versailles Saint-Quentin. He joined the PDI team in October 2024 in the goal of developing software to bridge between physical simulation, data analysis, and IA in the context of high-performance computing. Jacques obtained a PhD in applied mathematics and scientific computing from the University of Bordeaux I. Before joining the PDI team, he worked on software development for physical applications (for example FreeFEM).

Julian Auriac. CEA/MdlS

Julian Auriac is Research Engineer in computer science, HPC & AI, currently working at the Maison de la Simulation (MdlS), a french collaborative laboratory involving CEA, CNRS, and UVSQ. He is the latest addition to the PDI team, where his work focuses on PDI core and the DEISA plugin through the lenses of high-performance computing and parallel software engineering. After his engineering school, Julian specialized in HPC & AI through a specialized master’s course.

Julien Bigot, PhD . CEA/MdlS

Julien Bigot is a permanent CEA computer scientist at Maison de la Simulation where he leads the Science of Computing team. His research focuses on programming models for high-performance computing. He is especially interested in the question of separation of concerns between the simulated domain specific aspects and optimization aspects. He aims to concretize his research into production-level software like the DDC, Deisa, or PDI libraries whose development he started himself, or Kokkos, a project of the Linux Foundation originating from the US DoE. He also aims to apply these libraries in production into real codes like the Gysela code for example. He leads the CExA project that contributes to the Kokkos ecosystem, and in NumPEx, that intends to build the software stack for the French Exascale machines. He co-leads the Exa-DoST work on I/O and data analysis libraries and tools. In the Exa-DI project, he co-leads the work on software packaging, deployment, and CI/CD, and in the Exa-Soft project, he takes part in the work on GPU programming models.

Yushan Wang, PhD . CEA/MdlS

Yushan Wang has been a permanent research engineer at Maison de la Simulation, CEA Saclay, since 2021. She earned her Ph.D. in Computer Science in 2015, specializing in optimization of computational fluid dynamics (CFD) simulations for heterogeneous computing environments. Before joining Maison de la Simulation, she worked as an engineer at the Laboratory for Climate and Environmental Sciences (LSCE) from 2016, where she played a key role in developing an I/O library for climate models, optimizing their computational resource usage for I/O. Soon after joining Maison de la Simulation, she began contributing to the PDI library, a tool that decouples I/O concerns from numerical simulations. Her research interests include high-performance computing (HPC) I/O workflows, particularly in-situ and in-transit data analysis and in-situ data visualization.

Acknowledgements

As part of the “France 2030” initiative, this work has benefited from a State grant managed by the French National Research Agency (Agence Nationale de la Recherche) attributed to the Exa-DoST project of the NumPEx PEPR program, reference: ANR-22-EXNU-0004.

Contact

Feel free to reach out !
Click this image to get in touch through our Slack channel
<- Click me
Scan me ->
Scan this QR code to get in touch through our Slack channel