Civil Engineering Reference
In-Depth Information
NR value of 43, it is appreciably slower than the multi-grid insertion if the grid construction
time is also taken into consideration.
A cluster distribution is locally very similar to the uniform distribution, and not many
elongated tetrahedra are expected in the triangulation. As points are drastically unevenly
distributed, the NP value for regular grid insertion was pretty high, which increased steadily
with the number of points inserted up to more than 50 for a cluster of 2 million points; as a
result, the regular grid insertion took 50% more CPU time compared to uniform distribu-
tion. On the other hand, the multi-grid and kd-tree insertions both recorded an excellent
performance of a marginal increase in the CPU time relative to what is needed for the uni-
form distribution. Nevertheless, the performance of the multi-grid insertion is still the best
overall if grid construction is taken into account.
Apart from the number of points inserted, the impact of various non-uniform point dis-
tributions for the three insertion schemes has also been studied. The CPU times for various
distribution patterns by the insertion schemes for the insertion of 2 million and 20 million
points are plotted, respectively, in Figures 8.65 and 8.66. For the insertion of 2 million
points, regular grid insertion took much more CPU time (four times more in the worst case
of line distribution) for non-uniform point distributions compared to uniform distributions,
90
RG_2M
MG_2M
Kd_2M
80
70
60
50
40
30
20
10
0
Uniform
Line
Diagonal
Ellipse
Spiral
Cluster
Point distributions
Figure 8.65
CPU time for various distributions of 2 million points.
350
300
250
200
150
MG_20M
100
Kd_20M
50
0
Uniform
Line
Diagonal
Ellipse
Spiral
Cluster
Point distributions
Figure 8.66
CPU time for various distributions of 20 million points.
