A portable and customizable profiling framework for java based on bytecode instruction counting

  • Authors:
  • Walter Binder

  • Affiliations:
  • Artificial Intelligence Laboratory, Ecole Polytechnique Fédérale de Lausanne (EPFL), Lausanne, Switzerland

  • Venue:
  • APLAS'05 Proceedings of the Third Asian conference on Programming Languages and Systems
  • Year:
  • 2005

Quantified Score

Hi-index 0.00

Visualization

Abstract

Prevailing profilers for Java, which rely on standard, native-code profiling interfaces, are not portable, give imprecise results due to serious measurement perturbation, and cause excessive overheads. In contrast, program transformations allow to generate reproducible profiles in a fully portable way with significantly less overhead. This paper presents a profiling framework that instruments Java programs at the bytecode level to build context-sensitive execution profiles at runtime. The profiling framework includes an exact profiler as well as a sampling profiler. User-defined profiling agents can be written in pure Java, too, in order to customize the runtime processing of profiling data.