| | 14 | |
| | 15 | == Why does ANUGA get so slow when meshes are refined == |
| | 16 | |
| | 17 | Two reasons: |
| | 18 | |
| | 19 | * With smaller triangles timesteps tend to become smaller due to the CFL condition |
| | 20 | * The computation time per timestep grows more than linearly with the number of triangles. Here is why |
| | 21 | |
| | 22 | {{{ |
| | 23 | Let h be the grid size. The number of cells (triangles) N will normally be proportional to h^{-2} |
| | 24 | The size of the timesteps will be proportional to the grid size, and so the number of timesteps is proportional to h^{-1} = N^{1/2} |
| | 25 | So the computational cost (solution time) will be proportional to |
| | 26 | (number of cells) x (no of timesteps) = N x N^{1/2} = N^{3/2} |
| | 27 | log (Time) = 3/2 log( N). |
| | 28 | |
| | 29 | So increasing the grid size from 100 cells to 1000 cells grows the computational time by 10^{3/2} (about 30). |
| | 30 | }}} |
| | 31 | |
| | 32 | |