Training title: Maximizing Performance in High-Performance Computing – Tools, Techniques, and Hands-On Sessions

9:30 – 10:30 Welcome & Introduction to CINECA & LEONARDO - Caterina Caravita
- CINECA infrastructure and practical information for LEONARDO users

11.00 - 12:30 Introduction to package manger tools in HPC – Massimiliano Culpo
- SPACK: A flexible package manager supporting multiple versions, configurations, platforms, and compilers

14.00 - 16:30 Hands On to the applications
- How to manage and compile the code using SPACK

9:30 - 12:30 Performance Portability for NVIDIA’s GPU - Matt Bettencourt
- How GPU Computing Works, best practices when porting
- How application performance and portability can be rigorously measured using NVIDIA profiling tool

14.00 - 16:30 Hands On to the application
- Example code: Challenges and solutions porting the code

9:30 - 10:30 Using HPC tool Darshan – Laura Bellentani

11:00 – 12:30 Optimization by DDN - Jean Thomas

"Toward a more comprehensive performance Analysis: Adding I/O to the performance equation."
Performance analysts tend to focus their efforts on tuning algorithms and optimizing the implementation. In modern systems, the performance driver is a correct exploitation of parallelism. This parallelization effort often is not applied to the I/O part of the application, resulting in a significant serialization overhead, even though the I/O subsystem is a parallel file system.

This workshop proposes raising awareness of the importance of I/O optimizations and will provide tools and methods to implement these optimizations.

Overview of I/O systems: I/O patterns, the good, the bad and the ugly
concept and presentation

14:00 – 16:30
- Spotting I/O issues: patterns, scalability
Hands-on session with Darshan

- Optimization techniques
hands-on session on re-implementation of I/O patterns, using the MPI-IO configuration