Zoltan User's Guide  |  Next  |  Previous

Geometric (Coordinate-based) Partitioners

Geometric partitioners divide data into parts based on the physical coordinates of the data. Objects assigned to a single part tend to be physically close to each other in space. Such partitioners are very useful for applications that don't have explicit connectivity information (such as particle methods) or for which geometric locality is important (such as contact detection). They are also widely used in adaptive finite element methods because, in general, they execute very quickly and yield moderately good partition quality.

The geometric methods are the easiest non-trivial partitioners to incorporate into applications, as they require only four callbacks: two returning object information and two returning coordinate information.

We group refinement-tree partitioning for adaptive mesh refinement applications into the geometric partitioners because it uses geometric information to determine an initial ordering for coarse elements of adaptively refined meshes. The refinement-tree partitioner also requires tree-based callbacks with connectivity information between coarse and fine elements in refined meshes.

Recursive Coordinate Bisection (RCB)
Recursive Inertial Bisection (RIB)
Hilbert Space-Filling Curve Partitioning (HSFC)
Refinement Tree Based Partitioning (Reftree)

[Table of Contents  | Next:  Recursive Coordinate Bisection  |  Previous:  Random Partitioning  |  Privacy and Security]