Publicação
Data structure centered SPH performance evaluation
| Resumo: | Smoothed-particle Hydrodynamics (SPH) is a particle-based simulation considered by many to be the main candidate for fluid simulation. This model was developed by R.A. Gingold and J.J. Monaghan in 1977 and had the purpose of solving astrophysical problems. Over the years, Monaghan has revisited SPH (1985, 1988, 1992 and so on) and it also gained traction with other researchers who discovered new applications for the model such as ballistics, volcanology, oceanography, and so on. Among the fields there is one we are particularly interested in, and that is Fluid Simulation. This work aims to implement SPH using efficient data structures that allow a real-time simulation to run on the Graphics Processing Unit (GPU). According to the literature, the z-order indexing method and the hash map are the most suitable structures for this purpose. It is intended to see its impact and in which situations one will be better suited to use than the other. With said implementation, several tests were performed in order to analyze the robustness and stability of the method. With these tests it was possible to compare the two data structures used. The implemented SPH showed realistic and robust results in most cases, being able to handle multiple scenes of varying levels of complexity. Despite the good results, it showed some difficulties in maintaining stability in some boundaries (boundaries with great curvature or sharp edges) and also showed some difficulties in scenes with two fluids with different densities. As for the data structures, it was possible to observe that both are efficient and support real-time simulations with more than 1 million particles (using a NVIDIA RTX 3080). In the case of z-order, it proved to be the method with the best performance when compared to the hash map under the same conditions, that is, scenes with the same number of particles and the same simulation volume. This is due to the larger data locality that z-order has. On the other hand the hash map was a bit slower (when compared with the z-order under the same conditions) but allowed for greater freedom when creating a scene. When comparing the two methods with the same number of particles but different simulation volumes we can see that the hash map catches up with the z-order method as the particles spread across the simulation. With the two data structures analyzed it is possible to draw some conclusions. The z-order method is recom mended when we have a limited and relatively small simulation volume. In case there is no simulation volume, or it is very large, it is recommended to use a hash map since the performance deficit seems to disappear as the simulation volume gets bigger and the particles spread across the volume. |
|---|---|
| Autores principais: | Barbosa, Paulo Alexandre Ferreira |
| Assunto: | Fluid simulation Smoothed-particle hydrodynamics SPH Particle-based method Hash map Z-order Data structures Simulação de fluidos Método baseado em partículas Estruturas de dados Engenharia e Tecnologia::Engenharia Eletrotécnica, Eletrónica e Informática |
| Ano: | 2022 |
| País: | Portugal |
| Tipo de documento: | dissertação de mestrado |
| Tipo de acesso: | acesso aberto |
| Instituição associada: | Universidade do Minho |
| Idioma: | inglês |
| Origem: | RepositóriUM - Universidade do Minho |
| Resumo: | Smoothed-particle Hydrodynamics (SPH) is a particle-based simulation considered by many to be the main candidate for fluid simulation. This model was developed by R.A. Gingold and J.J. Monaghan in 1977 and had the purpose of solving astrophysical problems. Over the years, Monaghan has revisited SPH (1985, 1988, 1992 and so on) and it also gained traction with other researchers who discovered new applications for the model such as ballistics, volcanology, oceanography, and so on. Among the fields there is one we are particularly interested in, and that is Fluid Simulation. This work aims to implement SPH using efficient data structures that allow a real-time simulation to run on the Graphics Processing Unit (GPU). According to the literature, the z-order indexing method and the hash map are the most suitable structures for this purpose. It is intended to see its impact and in which situations one will be better suited to use than the other. With said implementation, several tests were performed in order to analyze the robustness and stability of the method. With these tests it was possible to compare the two data structures used. The implemented SPH showed realistic and robust results in most cases, being able to handle multiple scenes of varying levels of complexity. Despite the good results, it showed some difficulties in maintaining stability in some boundaries (boundaries with great curvature or sharp edges) and also showed some difficulties in scenes with two fluids with different densities. As for the data structures, it was possible to observe that both are efficient and support real-time simulations with more than 1 million particles (using a NVIDIA RTX 3080). In the case of z-order, it proved to be the method with the best performance when compared to the hash map under the same conditions, that is, scenes with the same number of particles and the same simulation volume. This is due to the larger data locality that z-order has. On the other hand the hash map was a bit slower (when compared with the z-order under the same conditions) but allowed for greater freedom when creating a scene. When comparing the two methods with the same number of particles but different simulation volumes we can see that the hash map catches up with the z-order method as the particles spread across the simulation. With the two data structures analyzed it is possible to draw some conclusions. The z-order method is recom mended when we have a limited and relatively small simulation volume. In case there is no simulation volume, or it is very large, it is recommended to use a hash map since the performance deficit seems to disappear as the simulation volume gets bigger and the particles spread across the volume. |
|---|