def run(self):
"""
09-05-05
Watch: when sending via MPI, tag 0 means from node 0, tag 1 means goes to the last node.
10-21-05
replace output_node() with the one from codense.common for better scheduling
--fill_edge2encodedOccurrence()
--input_node()
--get_cluster_block()
--computing_node()
--node_fire()
--output_node()
--output_cluster()
--uniqueSort()
"""
communicator = MPI.world.duplicate()
node_rank = communicator.rank
intermediateFile = "%s.unsorted" % self.outputfile # intermediateFile to store concatenated results
if communicator.rank == (communicator.size - 1):
edge2encodedOccurrence = {}
no_of_datasets = self.fill_edge2encodedOccurrence(
self.hostname, self.dbname, self.schema, edge2encodedOccurrence, self.min_sup, self.max_sup
)
mpi_synchronize(communicator)
if node_rank == 0:
self.input_node(
communicator, self.inputfile, self.min_size, self.cluster_block_size, self.cluster_block_edges
)
elif node_rank <= communicator.size - 2: # exclude the last node
self.computing_node(communicator, self.cluster_block_size, self.min_size, self.min_con)
elif node_rank == communicator.size - 1:
codense2db_instance = codense2db()
free_computing_nodes = range(1, communicator.size - 1)
writer = csv.writer(open(intermediateFile, "w"), delimiter="\t")
parameter_list = [writer, codense2db_instance, edge2encodedOccurrence, no_of_datasets]
output_node(
communicator,
free_computing_nodes,
parameter_list,
self.output_cluster,
report=self.report,
type=Numeric.Int,
)
del writer
# 10-21-05self.output_node(communicator, intermediateFile, codense2db_instance, edge2encodedOccurrence, no_of_datasets)
mpi_synchronize(communicator)
# collecting
if node_rank == 0:
MpiFromDatasetSignatureToPattern_instance = MpiFromDatasetSignatureToPattern()
MpiFromDatasetSignatureToPattern_instance.uniqueSort(intermediateFile, self.outputfile)
