The potential of sampling for dynamic analysis

  • Authors:
  • Joseph L. Greathouse;Todd Austin

  • Affiliations:
  • University of Michigan;University of Michigan

  • Venue:
  • Proceedings of the ACM SIGPLAN 6th Workshop on Programming Languages and Analysis for Security
  • Year:
  • 2011

Quantified Score

Hi-index 0.00

Visualization

Abstract

This paper presents an argument for distributing dynamic software analyses to large populations of users in order to locate bugs that cause security flaws. We review a collection of dynamic analysis systems and show that, despite a great deal of effort from the research community, their performance is still too low to allow their use in the field. We then show that there are effective sampling mechanisms for accelerating a wide range of powerful dynamic analyses. These mechanisms reduce the rate at which errors are observed by individual analyses, but this loss can be offset by the subsequent increase in test population. Nevertheless, there are unsolved issues in this domain that deserve attention if this technique is to be widely utilized.