This repository covers the I/O profiling training examples showcased at the OLCF User Summit 2026. The examples highlight good and bad I/O practices particulary regarding Lustre filesystems like Orion. Here we use C++ MPI-IO code to generate I/O profiles, but it's important to consider the workload that's occurring in these examples and extrapolate the ideas to non-MPI workloads such as AI I/O performance.
It's assumed you are working on a Cray system like Frontier and are using Frontier-related modules. To build the example code, run:
CC -o lustre_io lustre_lesson.cpp
Most I/O profiles found in io_profiles can be opened without any specific tools. The .map profile require the Linaro-Forge client
v 25+. The .darshan file requires the darshan-parser tool.
Examples of how to build the PyDarshan environment, and generate Darshan or Linaro-Forge MAP profiles can be found in the utils folder.