Domain Decomposition
|
|
Domain Decomposition |
|
USS_U2NCLE provides a built-in grid partitioning mechanism for domain decomposition of the grid into a specified number of subdomains. The partitioning process is fully automated and optimizes the load balance while at the same time minimizing the number of edges in the domain that are cut (span from one domain to another).
The number of subdomains must match the number of processors on which the solver run will be mapped. In this edit field, simply enter the number of subdomains.
Commonly used options for the grid decomposition process are listed in the "Options" listbox, and the individual options may be selected as needed. Advanced users may directly type command line options for the grid decomposition in the "Extra Options" edit field. Note that options typed in this field are not checked for correctness and could cause problems in the partitioner.
|
No Node Reordering
|
This option suppresses the reordering of the vertices of each subdomain. By default, vertex reordering is performed. The use of this option is discouraged, since the performance of the flow solver will be degraded by the nonlocality of the vertex ordering. |
|
No Element Sorting
|
This option suppresses sorting of the elements for the purpose of cache miss minimization. By default, element sorting is performed. The use of this option is discouraged, since the performance of the flow solver will be degraded by the nonlocality of the node-in-element ordering. |
|
Reduced Memory Footprint
|
Compress internal data structures temporarily whenever possible. This trades speed of the domain decomposition process for a lower memory usage footprint. |
|
Write Partitioning Vector Only
|
After partitioning, write the partitioning vector to a file named CaseName.pv and exit. This partitioning vector can then be used via the Read Partitioning Vector Only option described below. This pair of options is most useful in a severely memory constrained environment. |
|
Read Partitioning Vector Only
|
Instead of partitioning, read the partitioning vector from the file CaseName.pv. This file must have been previously created using the Write Partitioning Vector Only option. This pair of options is most useful in a severely memory constrained environment. |
Select the machine on which the grid decomposition should take place in the "Host" edit field. Start the grid decomposition by clicking "Launch".
For further information related to the materials in this web site, USS_U2NCLE, U2NCLE, SolidMesh, DIVA or information related to their use, please contact:
David L. Marcum
marcum@erc.msstate.edu
Phone: (662) 325-3193,
FAX: (662) 325-7692
Computational Simulation and Design Center
Engineering
Research Center for Computational Field Simulation
Mississippi State University
Box 9627, Mississippi State, MS
39762
Accesses: 368