Recording the control flow of parallel applications to determine iterative and phase-based behavior

  • Authors:
  • Karl Fürlinger;Shirley Moore

  • Affiliations:
  • Computer Science Division, EECS Department, University of California at Berkeley, Soda Hall 593, Berkeley, CA 94720, USA;Innovative Computing Laboratory, EECS Department, University of Tennessee, Claxton Complex, Knoxville, TN 37996, USA

  • Venue:
  • Future Generation Computer Systems
  • Year:
  • 2010

Quantified Score

Hi-index 0.01

Visualization

Abstract

Many applications exhibit iterative and phase-based behavior. We present an approach to detect and analyze iteration phases in applications by recording the execution control flow graph of the application and analyzing it for loops that represent iterations. Phases are then manually marked and performance profiles are captured in alignment with the iterations. By analyzing how profiles change between capture points the differences in execution behavior between iterations can be highlighted.