Algorithm Overview

The block partitioning method is a simple partitioning method that partitions objects into groups of objects with consecutive IDs, where each group is assigned to the same part. This partitioning is computed so that the sum of the weights of the objects on each part is balanced (for unweighted objects, the number of objects is balanced).

Input

The BlockPartitioning method expects a Zoltan2::IdentifierInput object that will contain the IDs of the objects and optionally object weights.

Parameters

In order to enable Block:

algorithm should be set to block

The following parameters are used by the Block algorithm:

imbalance_tolerance (default is 1.1 )
partitioning_objective (default: if weights specified, balance_object_weights; otherwise balance_object_count)

Solution

The solution is a list of part assignments for each object and the load imbalance for each weight.

Quality measures

Partitioning quality is measured with an imbalance measure. Use the parameter compute_metrics if you want the Zoltan2::PartitioningProblem to compute imbalance metrics for the solution.

Examples

block.cpp

Source

Zoltan2_AlgBlock.hpp is the source file for block partitioning.