AMROC Main     Blockstructured Adaptive Mesh Refinement in object-oriented C++

Periodic boundary conditions - 3 Grid Levels   (density, isosurface at 1.2)

Problem description       Results:   3 Levels
previous    



 
t=0
t=0.21
 
t=0.84
  


Analogue to 2D computation (cuts at z=0.0 and z=0.5)


Benchmark

 
 Task
P=1
P=2
P=4, BufferWidth=2
P=4, BufferWidth=1
P=4, Uniform
 
s
%
s
%
s
%
s
%
s
%
Integration
 47865
  91.6
 24065
  78.4
 12405
  72.8
  7363
  64.9
 13212
  97.9
Flux correction
   799
   1.5
  1368
   4.5
  1116
   6.6
     0
   0.0
     0
   0.0
Boundary setting
   530
   1.0
  2709
   8.8
  1926
  11.3
  1720
  15.2
   129
   1.0
Recomposition
  2364
   4.5
  2091
   6.8
  1326
   7.8
  1724
  15.2
     0
   0.0
Clustering
   415
   0.8
   225
   0.7
   114
   0.7
    51
   0.4
     0
   0.0
Misc.
   262
   0.5
   222
   0.8
   133
   0.8
   480
   4.3
   148
   1.2
Total / Parallel Efficiency
 52235
 100.0
 30679
  85.1
 17021
  76.7
 11338
 115.2
 13489
  96.8



In this example large regions of the computational domain require adaption. An uniformly refined calculation needs less computational time on four computing nodes (P=4) than an adaptive computation with the usual buffer witdth of two additional cells. If the buffer width is reduced to one cell, the adaptive computation catches up. But, the calculated result becomes incorrect. This example shows drastically that in the general case a buffer width of two cells is unavoidable.

Comparison between the two adaptive and the uniformly refined computation

top

Quickstart       Users Guide     Programmers Reference     Installation      Examples     Download


AMROC Main      Home      Contact
last update: 6/1/04