Zoltan User's Guide  |  Next  |  Previous

Ordering Algorithms

The following graph (sparse matrix) ordering algorithms are currently included in the Zoltan toolkit:
Nested dissection by METIS/ParMETIS
Nested dissection by Scotch/PT-Scotch
Local ordering with Hilbert space filling curves
These methods produce orderings for various applications (e.g., reducing fill in sparse matrix factorizations). Ordering is accessed through calls to Zoltan_Order.

Third-party libraries

Currently, most ordering in Zoltan is provided through the third-party libraries METIS/ParMETIS and PT-Scotch. The one exception is the local Hilbert space filling curve ordering. To use the other methods, a third-party library must be present.

Ordering Parameters

While the overall behavior of Zoltan is controlled by general Zoltan parameters, the behavior of each ordering method is controlled by parameters specific to ordering which are also set by calls to Zoltan_Set_Param. Many of these parameters are specific to individual ordering algorithms, and are listed in the descriptions of the individual algorithms.  However, several have meaning across multiple ordering algorithms. These parameters are described below.
 
Parameters:
   ORDER_METHOD The order algorithm used by Zoltan is specified by this parameter. Valid values are 
"METIS" (sequential nodal nested dissection by METIS),
"PARMETIS" (parallel nodal nested dissection by ParMETIS ),
"SCOTCH" (sequential ordering using Scotch),
"PTSCOTCH" (parallel ordering using PT-Scotch),
"LOCAL_HSFC" (local ordering using Hilbert space filling curves), and
"NONE" (for no ordering).
Default Values:
ORDER_METHOD = PARMETIS (with MPI), METIS (when no MPI)

[Table of Contents  | Next:  Nested dissection by ParMETIS  |  Previous:  Hybrid Hierarchical Partitioning  |  Privacy and Security]