SST::Partition::SSTLinearPartition Class Reference

Performs a linear partition scheme of an SST simulation configuration. More...

#include <linpart.h>

Inheritance diagram for SST::Partition::SSTLinearPartition:

Public Member Functions
	SSTLinearPartition (int mpiRankCount, int verbosity)
	Creates a new linear partition scheme.
void	performPartition (PartitionGraph *graph)
	Performs a partition of an SST simulation configuration.
bool	requiresConfigGraph ()
bool	spawnOnAllRanks ()
Static Public Member Functions
static SSTPartitioner *	allocate (int total_ranks, int my_rank, int verbosity)
Protected Attributes
int	rankcount
	Number of ranks in the simulation.
Output *	partOutput
	Output object to print partitioning information.
Static Protected Attributes
static bool	initialized = SSTPartitioner::addPartitioner("linear",&SSTLinearPartition::allocate, "Partitions components by dividing Component ID space into roughly equal portions. Components with sequential IDs will be placed close together.")

Detailed Description

Performs a linear partition scheme of an SST simulation configuration.

In this scheme a list of components (supplied as a graph) are grouped by slicing the list into approximately equal parts. A "part" is generated for each MPI rank performing the simulation. This means Components with sequential ids will be placed close together. In general this scheme provides a very effective partition for most parallel simulations which generate man similar components of interest close together in the input Python configuration. It is also very fast to compute a linear partiton scheme. For more aggressive partition schemes users should try either a simple or Zoltan-based partitioner.