Difference between revisions of "Quadtree and Sub-Grid Sampling FAQ"
| Line 7: | Line 7: | ||
Not necessarily. By default, running model on a mesh (Quadtree) rather than a grid (HPC) is always slower, on average 20%. Quadtree really comes into an effect once there is at least three levels of cell size and judicious refinement - around 80% of cell count reduction. As Quadtree development is an ongoing task this might further improve in the future. | Not necessarily. By default, running model on a mesh (Quadtree) rather than a grid (HPC) is always slower, on average 20%. Quadtree really comes into an effect once there is at least three levels of cell size and judicious refinement - around 80% of cell count reduction. As Quadtree development is an ongoing task this might further improve in the future. | ||
| − | =Why I should not use a | + | =Why I should not use a combination of SGS and Quadtree?= |
The main advantages of using Quadtree are shorter run times (with considerable cell count reduction) and smaller size of output files. If longer run times of model without Quadtree are not an issue and output file sizes are fine then there is no strong argument to use Quadtree. Some models with cell count reduction only around 30% might be even running slower with Quadtree than the original HPC model. | The main advantages of using Quadtree are shorter run times (with considerable cell count reduction) and smaller size of output files. If longer run times of model without Quadtree are not an issue and output file sizes are fine then there is no strong argument to use Quadtree. Some models with cell count reduction only around 30% might be even running slower with Quadtree than the original HPC model. | ||
Revision as of 15:13, 15 July 2020
This Page is under construction
How coarse can base cell size be for Quadtree model with Sub-Grid Sampling (SGS)?
Even when using SGS to improve conveyance, only one velocity is used for the whole cell. Every model has a cell size range, from very fine to very coarse, beyond which it just doesn't make any sense to go. Cell size sensitivity testing is recommended to establish this range.
Should the same model using Quadtree with smaller cell count be always faster than HPC?
Not necessarily. By default, running model on a mesh (Quadtree) rather than a grid (HPC) is always slower, on average 20%. Quadtree really comes into an effect once there is at least three levels of cell size and judicious refinement - around 80% of cell count reduction. As Quadtree development is an ongoing task this might further improve in the future.
Why I should not use a combination of SGS and Quadtree?
The main advantages of using Quadtree are shorter run times (with considerable cell count reduction) and smaller size of output files. If longer run times of model without Quadtree are not an issue and output file sizes are fine then there is no strong argument to use Quadtree. Some models with cell count reduction only around 30% might be even running slower with Quadtree than the original HPC model.