Internal:Ss missile comparison AFLR: Difference between revisions

From crtc.cs.odu.edu
Jump to navigation Jump to search
Ctsolakis (talk | contribs)
Ctsolakis (talk | contribs)
Line 79: Line 79:


= Qualitative =
= Qualitative =
== Starting From a Surface Mesh ==
[[File:AFLR_CDT3D_edges.png|500px]]
[[File:AFLR_CDT3D_edges.png|500px]]
[[File:AFLR_CDT3D_edges_log.png|500px]]
[[File:AFLR_CDT3D_edges_log.png|500px]]

Revision as of 08:23, 18 September 2018

Quantitative

Input : Surface mesh, background mesh and metric tensor on background mesh

Metric Complexity # Vertices # Elements Time (s)
Target 80,000 ~160,000 ~960,000 N/A
AFLR (without metric alignment) 80,299.6 130,668 744,285 341.92
AFLR (with metric alignment) 80,422.6 127,112 732,883 238.57
CDT3D (1 thread) 78,215.9 189,788 1,107,104 100.0
CDT3D (6 threads) 79,221.6 191,353 1,116,112 21.5

Input : background mesh and metric tensor on background mesh (tets 590,102 , vertices 103,781)

Metric Complexity # Vertices # Elements Time (s)
AFLR (without metric alignment)
AFLR (with metric alignment) 79,666.1 116,024 664,501 15.130
CDT3D (1 thread) 79,333.4 126,877 727,840 38.134
CDT3D (2 threads) 79,314.6 126,968 728,240 21.01
CDT3D (4 threads) 79,312.3 127,207 729,640 12.38

Qualitative

Starting From a Surface Mesh

Time Breakdown

Operations details AFLR (malign =ON) AFLR(maling =OFF)
BG Mesh Initialization 0.534 0.533
Point Creation
Euclidean length 0.028 0.037
Riemannian length 0.166 0.24
Metric interpolation at face 0.302 0.408
Evaluate Log(Metric) 1.902 0.272
Search in BG mesh 137.825 334.594
Search point in mesh 0.092 0.271
Metric-Alignment 80.387 0
Point Creation Total 219.759 336.881
Point Filtering 2.373 2.412
Point Insertion 0.154 0.184
Element Deactivation 0.626 2.914
Local Reconnection 32-23 8.553 9.037
Local Reconnection 4-4 0 3.118
Total 232.389 355.571
Operations CDT3D ( 1 thread )
BG Mesh Initialization 1.08535
Deactivate Elements 2.69843
Point Creation 8.54597
Point Insertion 0.25458
Local Reconnection 3-2/2-3 64.4204
Local Reconnection 4-4 22.4864
Total 100.058

Comments

  1. Although CDT3D does not use alignment, it is not this that makes it faster (actually when alignment will be implemented, it is expected to make refinement even faster)
  2. By design, the point insertion order in CDT3D favours points that are geometrically close in space.
  3. In CDT3D, we store the previous location in BG mesh, speeding up subsequent queries in the same neighborhood.