Simulation of Task Graph Systems in Heterogeneous Computing Environments

This paper describes a simulation tool for the analysis of complex jobs described in the form of task graphs. The simulation procedure relies on the PN-based topological representation of the task graph that takes advantage of directly modeling precedence constraints and other characteristics inherent in Generalized Stochastic Petri Nets (GSPN). The GSPN representation is enhanced with enabling functions that govern the sequence of firings of transitions representing execution of tasks. The regulated flow of activity is carried out observing not only precedence constraints but specific allocation heuristics and communication delays. The tool is useful in evaluating different heuristics described by the corresponding implemented algorithm, or using a deterministic timespan given by a Gantt chart.