LARGE-SCALE MULTITHREADING SELF-GRAVITY SIMULATIONS FOR ASTRONOMICAL AGGLOMERATES

This article presents parallel multithreading self-gravity simulations for astronomical agglomerates, applying High Performance Computing techniques to allow the efficient simulation of systems with a large number of particles. Considering the time scales needed to properly simulate the processes involved in the problem, two parallel mesh-based algorithms to speed up the self-gravity calculation are proposed: a method that updates the occupied cells of the mesh, and a method to divide the domain based on the Barnes-Hut tree. Results of the experimental evaluation performed over a scenario with two agglomerates orbiting each other indicate that the Barnes-Hut allows accelerating the execution times over 10x compared to the occupied cells method. These performance improvements allow scaling up to perform realistic simulations with a large number of particles (i.e., tens of millions) in reasonable execution times.