Highly scalable multi objective test suite minimisation using graphics cards

  • Authors:

  • Shin Yoo;Mark Harman;Shmuel Ur

  • Affiliations:

  • University College London;University College London;University of Bristol

  • Venue:
  • SSBSE'11 Proceedings of the Third international conference on Search based software engineering
  • Year:
  • 2011

Quantified Score

Hi-index 0.00

Visualization

Abstract

Despite claims of "embarrassing parallelism" for many optimisation algorithms, there has been very little work on exploiting parallelism as a route for SBSE scalability. This is an important oversight because scalability is so often a critical success factor for Software Engineering work. This paper shows how relatively inexpensive General Purpose computing on Graphical Processing Units (GPGPU) can be used to run suitably adapted optimisation algorithms, opening up the possibility of cheap scalability. The paper develops a search based optimisation approach for multi objective regression test optimisation, evaluating it on benchmark problems as well as larger real world problems. The results indicate that speed-ups of over 25x are possible using widely available standard GPUs. It is also encouraging that the results reveal a statistically strong correlation between larger problem instances and the degree of speed up achieved. This is the first time that GPGPU has been used for SBSE scalability.