Skip to content

Decomposition · Changes

Page history
Updated Decomposition (markdown) authored by rundeck's avatar rundeck
Hide whitespace changes
Inline Side-by-side
Decomposition.md
View page @ 1652d0eb
...@@ -21,19 +21,7 @@ The Fixed decomposition just divide the domain as the user specified and assign ...@@ -21,19 +21,7 @@ The Fixed decomposition just divide the domain as the user specified and assign
The data driven decomposition instead consider the volume of the information the structure store and divide the volume equally them equally regardless of the communication. The data driven decomposition instead consider the volume of the information the structure store and divide the volume equally them equally regardless of the communication.
### Model decomposition ### Model decomposition
* A
* B
<div id="sliderFrame">
<div id="slider">
<img src="uploads/images/domain.jpg" alt="Initial domain" />
<img src="uploads/images/domain_decomposed.jpg" alt="Divide the space into a grid of sub-sub-domain" />
<img src="uploads/images/domain_graph.jpg" alt="Generate the graph model" />
<img src="uploads/images/domain_graph_decomposed.jpg" alt="Find optimal decomposition" />
<img src="uploads/images/domain_subdomain_decomposed.jpg" alt="Post process the decomposition" />
<img src="uploads/images/domain_subdomain_decomposed_wg.jpg" alt="Sub-domain creation" />
</div>
</div>
In a distributed-memory setting, where data are scattered across In a distributed-memory setting, where data are scattered across
processors, two factors are important: equal division of work across processors, two factors are important: equal division of work across
...@@ -50,6 +38,17 @@ weights, and the sum of the cut edges is minimal (_Find optimal decomposition_). ...@@ -50,6 +38,17 @@ weights, and the sum of the cut edges is minimal (_Find optimal decomposition_).
the decomposition can be post-process further more to be more optimal and based on factors not considered by the optimization process, like merging vertex, or merging sub-sub-domain, to create bigger sub-domains. the decomposition can be post-process further more to be more optimal and based on factors not considered by the optimization process, like merging vertex, or merging sub-sub-domain, to create bigger sub-domains.
Even if a model decomposition it is not bind to a graph model it is true that until now it is the best approach in scientific computation Even if a model decomposition it is not bind to a graph model it is true that until now it is the best approach in scientific computation
<div id="sliderFrame">
<div id="slider">
<img src="uploads/images/domain.jpg" alt="Initial domain" />
<img src="uploads/images/domain_decomposed.jpg" alt="Divide the space into a grid of sub-sub-domain" />
<img src="uploads/images/domain_graph.jpg" alt="Generate the graph model" />
<img src="uploads/images/domain_graph_decomposed.jpg" alt="Find optimal decomposition" />
<img src="uploads/images/domain_subdomain_decomposed.jpg" alt="Post process the decomposition" />
<img src="uploads/images/domain_subdomain_decomposed_wg.jpg" alt="Sub-domain creation" />
</div>
</div>
* CartDecomposition [(doc)](CartDecomposition)[(API reference)](http://ppmcore.mpi-cbg.de/doxygen/openfpm_pdata/classCartDecomposition.html) * CartDecomposition [(doc)](CartDecomposition)[(API reference)](http://ppmcore.mpi-cbg.de/doxygen/openfpm_pdata/classCartDecomposition.html)
### Fixed decomposition ### Fixed decomposition
......