Parallel Performance Visualization: From Practice to Theory

  • Authors:
  • Michael T. Heath;Allen D. Malony;Diane T. Rover

  • Affiliations:
  • -;-;-

  • Venue:
  • IEEE Parallel & Distributed Technology: Systems & Technology
  • Year:
  • 1995

Quantified Score

Hi-index 0.00

Visualization

Abstract

Parallel performance visualization presents fundamental challenges--complex data sets, the artificial aspects of parallel computation, the need for integrated analysis and visual models, and the dependence on the user's mental perspective. Despite these challenges, important concepts for visualization design and use have begun to emerge.A history of the practice of implementing and using visualization in performance analysis tools reveals an underlying theory in the form of an abstract model of performance visualization and applied visualization concepts. The model is based on the integration of performance evaluation models and performance displays. The concepts involve aspects of context, scaling, user perception and interaction, comparison, and extraction of information.Several performance scenarios demonstrate how to apply the model and concepts and principles in real tools to overcome the problems of understanding performance. These scenarios measure utilization and communication, utilization statistics, data access, and parallel performance optimization. The model and concepts form a solid basis for future research in parallel performance visualization.