GSPN subnet aggregation algorithms for modeling distributed applications in parallel architectures

This paper presents algorithms for the mean response time of generalized stochastic Petri nets (GSPN) models based on a methodology for performance modeling and prediction of real time distributed applications running in parallel architectures. For architectures with monoprogrammable processors, the mean response time is obtained by successive aggregation without the use of state space, which prevents state explosion. For multiprogrammable processor architectures, reductions in immediate transitions are applied. The methodology described herein aims to help the designer to predict application execution times early in their developmental phase.