From Data Traces to System Insight—Redefining I/O Understanding in HPC

Speaker: Prof. Dr. Sarah Neuwirth, Full Professor of Computer Science, Johannes Gutenberg University Mainz

Title: From Data Traces to System Insight—Redefining I/O Understanding in HPC

Slides: https://hpc.fau.de/files/2025/07/2025-NHR-PerfLab-Talk-Neuwirth.pdf

Abstract:
As high-performance computing systems evolve to accommodate increasingly heterogeneous and data-intensive workloads, I/O behavior is emerging as a critical determinant of system efficiency, reliability, and scalability. This talk explores how rethinking I/O as a source of insight, rather than simply a performance constraint, opens new pathways for analysis, optimization, and design. I will discuss the growing complexity of I/O patterns across modern workflows, the challenges of achieving observability across deeply layered I/O stacks, and the importance of trace-driven methodologies for ensuring reproducibility and explainability.

We will examine how these factors influence performance variability, system design trade-offs, and future benchmarking practices, particularly in the context of shifting architectural paradigms, emerging consistency models, and the increasing role of adaptive, multi-tiered storage. The goal of this talk is to offer a perspective on I/O that is both system-aware and forward-looking, highlighting its central role in shaping the next generation of sustainable and transparent HPC infrastructures.

For a list of past and upcoming NHR PerfLab seminar events, see: https://hpc.fau.de/research/nhr-perflab-seminar-series/