Systematic generation and evaluation of stochastic petri net models for the performance analysis of task graphs

The main thrust of this work is to develop a mechanism to generate a Stochastic Petri Net from a given task graph. The task graph corresponds to a particular job that has been broken down into several tasks. The problem is to find the average completion time of any job represented by a task graph using an unlimited number of processors as v^ell as a limited number of processors. Currently, there are models that solve the problem for an unlimited number of processors [10]. However, there is no model available to solve this problem for a limited number of processors. Work has been done using Queuing Networks [13], but not Stochastic Petri Nets. To analyze a particular task graph using a Petri Net with a limited number of processors, the Stochastic Petri Net(SPN) model must be generated manually. Once the Stochastic Petri Net is generated, its evaluation can be carried out using a tool such as the Stochastic Petri Net Package (SPNP) [1] developed at Duke University.