Identification of performance characteristics from multi-view trace analysis

  • Authors:
  • Daniel Spooner;Darren Kerbyson

  • Affiliations:
  • Dept. Of Computer Science, University of Warwick, Coventry, UK;Performance and Architectures Laboratory, CCS-3, Los Alamos National Laboratory, Los Alamos

  • Venue:
  • ICCS'03 Proceedings of the 2003 international conference on Computational science: PartIII
  • Year:
  • 2003

Quantified Score

Hi-index 0.00

Visualization

Abstract

In this paper, we introduce an instrumentation and visualisation tool that can be used to assist in analytical performance model generation. It is intended to provide a means of focusing the interest of the performance specialist, rather than automating the entire formulation process. The key motivation for this work was that while analytical models provide a firm basis for conducting performance studies, they can be time-consuming to generate for large, complex applications. The tool described in this paper allows trace files from different runs of an application to be compared and contrasted in order to determine the relative performance characteristics for critical regions of code. It is envisaged that the tool will develop to identify and summarise specific performance issues such as communication strategies through the use of novel visualisation techniques.