Multi-Threading and GPU Acceleration: The Case for Faster Simulators for Maritime Search and Rescue Operations Planning

Abstract

In decision support systems (DSSs) for maritime search and rescue (MSAR), simulation can be used to calculate the probability of success of a search operation tasking aircraft and vessels to search patterns. Nonetheless, single-threaded simulators remain a bottleneck in DSSs for MSAR, where thousands of candidate search operations generated by an optimizer have to be evaluated within a few minutes. Slow simulations also influence the design of DSSs when slight manual corrections to an operation require waiting for reevaluation. In response to the need for faster simulators, we redesigned the simulation pipeline and developed two parallel simulators: one based on CPU multi-threading and one based on GPU acceleration. Experiments show that CPU multi-threading provides a 6.7-fold acceleration over the single-threaded baseline, whereas the GPU acceleration yields a 260-fold speedup. Such speedups can render DSSs more responsive and enable optimizers to provide better recommendations, ultimately leading to more effective MSAR operations.