Thesis title: On Techniques to Handle Risk in Speculative Parallel Discrete-Event Simulation
In this thesis, the traditional concepts of risk and aggressiveness associated with speculative parallel/distributed execution of discrete-event simulations are revisited and re-examined, introducing a more general framework of interpretation. It is shown how these concepts, intrinsic to the speculative nature of parallel/distributed executions, can and should be addressed in competitive simulation engine implementations. The spectrum of possibilities related to the exploitation of risk and aggressiveness is explored from both a methodological and a technological point of view, on which strong emphasis is placed. We illustrate techniques that improve performance while providing several guarantees to model developers. In particular, it is shown how it is possible to combine different synchronisation algorithms to avoid thrashing phenomena and provide transparent support to the programmer that avoids known problems or significant performance drops. In general, revisiting the concepts of risk and aggressiveness opens up the possibility of repositioning concepts from the literature into a broader utilisation framework, which can be exploited in future lines of research to produce increasingly high-performance simulation systems.