High Performance Asynchronous I/O for Exascale Spectral Element Methods
Inside Livermore Lab via YouTube
Get 20% off all career paths from fullstack to AI
Power BI Fundamentals - Create visualizations and dashboards from scratch
Overview
AI, Data Science & Cloud Certificates from Google, IBM & Meta — 40% Off
One plan covers every Professional Certificate on Coursera. 40% off Coursera Plus Annual.
Unlock All Certificates
This seminar from the FEM@LLNL Series features Dr. Freddie Witherden of Texas A&M University discussing "High Performance Asynchronous I/O for Exascale Spectral Element Methods." Explore how modern simulations face challenges with disk I/O lagging behind memory bandwidth and arithmetic capabilities on leadership-class machines, resulting in disproportionate time spent writing checkpoint files. Discover the innovative massively parallel asynchronous file format debuting with PyFR v3, specifically designed for discontinuous spectral element methods. Learn about its key features including space-efficient node-based mesh description, compact connectivity arrays for fast parallel interface construction, self-describing nodal solution representations, support for partial solutions, multiple pre-computed partitionings, and partitioning-independent output files—all with entirely contiguous disk I/O patterns. The presentation explains how this HDF5-based format with custom I/O routines enables more efficient parallel I/O and asynchronous capabilities, concluding with insights on deployment benefits, particularly in bypassing outdated vendor-provided HDF5 builds. Sponsored by the MFEM project, this hour-long talk is part of a series focused on finite element research and applications of interest to the MFEM community.
Syllabus
FEM@LLNL | High Performance Asynchronous I/O for Exascale Spectral Element Methods
Taught by
Inside Livermore Lab