Design and Evaluation of Scalable Representations of Communication in Gantt Charts for Large-scale Execution Traces

TL;DR

This paper introduces scalable visual representations for communication structures in large-scale HPC execution traces, improving interpretability in dense Gantt charts through glyph-based visualization.

Contribution

The paper designs and evaluates glyph-based metaphors for visualizing communication patterns in large-scale HPC traces, addressing scalability issues in traditional Gantt charts.

Findings

Glyph representations improve pattern recognition accuracy.

Users interpret communication structures more effectively with the new visualization.

The approach scales to thousands of processors without losing clarity.

Abstract

Gantt charts are frequently used to explore execution traces of large-scale parallel programs found in high-performance computing (HPC). In these visualizations, each parallel processor is assigned a row showing the computation state of a processor at a particular time. Lines are drawn between rows to show communication between these processors. When drawn to align equivalent calls across rows, structures can emerge reflecting communication patterns employed by the executing code. However, though these structures have the same definition at any scale, they are obscured by the density of rendered lines when displaying more than a few hundred processors. A more scalable metaphor is necessary to aid HPC experts in understanding communication in large-scale traces. To address this issue, we first conduct an exploratory study to identify what visual features are critical for determining similarity between structures shown at different scales. Based on these findings, we design a set of glyphs for displaying these structures in dense charts. We then conduct a pre-registered user study evaluating how well people interpret communication using our new representation versus their base depictions in large-scale Gantt charts. Through our evaluation, we find that our representation enables users to more accurately identify communication patterns compared to full renderings of dense charts. We discuss the results of our evaluation and findings regarding the design of metaphors for extensible structures.