Speaker
Description
The complexity of the HPC I/O stack combined with gaps in the state-of-the-art profiling tools creates a barrier that does not help end-users and scientific application developers solve the I/O performance problems they encounter. To bridge this gap, we introduce Dristhi, a multi-source interactive analysis framework that empowers users to visualize I/O traces, identify bottlenecks, and gain a deeper understanding of application behavior. By combining cross-layer analysis with heuristic-based automatic detection, Dristhi provides actionable insights and recommendations to overcome common I/O performance bottlenecks. This talk will delve into the design and capabilities of Dristhi, demonstrate its use in pinpointing I/O performance issues, and highlight upcoming features that will further enhance its functionality.
