Missing the Trees for the Branches: Graphical-Scripting Interaction with Large-Scale Calling Context Trees

Student: Connor Scully-Allison (University of Arizona, Lawrence Livermore National Laboratory)
Supervisor: Katherine Isaacs (University of Arizona)

Abstract: In large scale performance analysis of HPC code, there is a tension between the visualizations provided by automated tools and the flexibility afforded by custom scripting. For calling context trees (CCT), node-and-link visualizations give users a view of performance data that corresponds to their mental model. Unfortunately, while current scripting workflows support the writing of code, they are not well supported by existing visualizations. GUI-based tools often restrict users to rigid modes of interaction. Furthermore, many general purpose visualization libraries are not well configured to support the complex data collected by HPC profilers, do not scale well to fit real HPC datasets, and often require exporting and viewing visualizations in a context divorced from where the code is developed. To ease this tension and support modern analysis workflows, we use human-centric design methodologies to marry scripting and graphical interaction paradigms into a scalable, interactive CCT visualization embedded in Jupyter notebooks.

ACM-SRC Semi-Finalist: no

Poster: PDF
Poster Summary: PDF


Back to Poster Archive Listing