Understanding parallel program behavior through petri net models

Generalized stochastic Petri nets (GSPN) are proposed in this paper as a way to represent and analyze programs written with process-oriented parallel languages. Following an approach that has already been proposed in the literature, parallel programs are translated into GSPN models that are used to perform a static analysis yielding information on possible deadlocks, mutual exclusion, and resource utilization. These models offer help in understanding the overall behavior of a parallel program, increase confidence in its correctness, and permit assessment of the efficiency of its implementation. The feasibility of automatic generation of Petri nets from program code is discussed and the problem of correctly modeling communications and flow control variables is addressed.