Simulation analysis of multithreaded programs under deadlock-avoidance control

We employ discrete event simulation to evaluate the performance of deadlock-prone multithreaded programs, either general-purpose software or parallel simulators, under a novel technique for deadlock-avoidance control recently proposed in the literature. The programs are modeled by a special class of Petri nets, called Gadara nets. We propose a formal simulation methodology for Gadara nets. We then use simulation to analyze two deadlock-prone multithreaded programs, where we study system performance in terms of safety, efficiency, and activity level, both before and after deadlock-avoidance control is applied. We further conduct a sensitivity analysis to investigate the effect of key parameters on the program's performance. We discuss the implications of the above results on the practical implementation of control strategies that prevent deadlocks in multithreaded programs.