Performance Evaluation and Concurrent Programming

Concurrent programs are developed to meet the demands of high performance computing of many new scientific and real time applications. Concurrent programming is however still difficult because of the lack of tools that help in developing and debugging new efficient implementations. Concurrency, communication, synchronization, and non- determinism make the manual assessment of the correctness and of the efficiency of concurrent programs extremely difficult. Models of concurrent programs must thus be used since the early stages of their development to support their debugging and tuning. In this paper we discuss the role that Performance Evaluation techniques may play within this environment using as an example a methodology that has been recently proposed for the automatic construction of GSPN models of concurrent applications. In particular, it will be shown that a GSPN model of an application can be directly derived from its code and that its evaluation provides the parameters that are needed for obtaining the optimal allocation of the components of the application on the computational units of a parallel architecture.