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


Backward facing step - 2 Grid Levels (density at time t = 0.75)

Problem description       Results:   2 Levels        3 Levels
previous     next



Contour lines of density on refinement grids (P=1)


 
Level 0
Level 1
Grids
    2
   52
Cells
67500
22556





Benchmark

Task
P=1
P=2, GuC-Factor=2
P=4, GuC-Factor=2
P=2, GuC-Factor=1
P=4, GuC-Factor=1
s
%
s
%
s
%
s
%
s
%
Integration
  332
 58.8
  170
 41.8
   81
 33.1
  218
 23.9
  104
 22.0
Flux correction
   41
  7.3
   27
  6.6
   15
  6.1
   31
  3.4
   19
  4.0
Boundary setting
   23
  4.1
   97
 23.8
   70
 28.6
   93
 10.2
   62
 13.2
Recomposition
   27
  4.1
   97
 23.8
   70
 28.6
  483
 53.0
  227
 48.1
Clustering
   40
  7.0
   19
  4.7
   10
  4.1
   19
  2.1
   10
  2.2
Misc.
  102
 18.0
   44
 10.8
   30
 12.2
   58
  7.4
   40
 10.5
Total / Parallel Efficiency
  565
100.0
  407
 69.4
  245
 57.7
  901
 31.4
  462
 30.6

If a relatively fine base grid is used, the choice of a larger GuC-Factor can decrease the overhead for partitioning and recomposition significantly. In this example, the partitioner can use up to 67500 units for GuC-Factor=1, while for GuC-Factor=2 only up to 16875 units are allowed.


top


Quickstart       Users Guide     Programmers Reference     Installation      Examples     Download



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