... | ... | @@ -20,6 +20,8 @@ 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.
|
|
|
|
|
|
### Domain definition
|
|
|
|
|
|
In the following we will consider structures that map over N-dimensional spaces because at the moment is our main interest
|
|
|
|
|
|
### Model decomposition
|
... | ... | @@ -28,7 +30,7 @@ In a distributed-memory setting, where data are scattered across |
|
|
processors, two factors are important: equal division of work across
|
|
|
processors and reduction of the communication overhead.
|
|
|
A typical approach, is to formulate the problem as a
|
|
|
graph-partitioning problem: the physical domain is divided into sub-sub-domains (_Domain decomposition_)
|
|
|
graph-partitioning problem: the domain is divided into sub-sub-domains (_Domain decomposition_)
|
|
|
(vertices of the graph), each of them carrying a weight modelling the computational cost.
|
|
|
The communication pattern between sub-domains is represented as links
|
|
|
between the sub-sub-domains (edges of the graph) with weights formalizing
|
... | ... | |