Performance analysis for distributed and parallel Java programs with Aksum

This paper deals with the challenging problem of performance analysis for Java programs. We describe procedures and requirements for instrumenting, monitoring, and analyzing distributed Java codes, and introduces Aksum, a highly customizable and flexible system for performance analysis that helps programmers to semi-automatically locate and understand performance problems in parallel and distributed Java programs. We also describe a sophisticated agent architecture as part of Aksum for static and dynamic instrumentation of Java programs. Experiments are presented for a widely distributed application running on a heterogeneous set of machines with different operating systems to illustrate the usefulness of our approach.