A hypergraph consists of vertices and hyperedges. A hyperedge connects one or more vertices. A graph is a special case of a hypergraph where each edge has size two (two vertices). The hypergraph model is well suited to parallel computing, where vertices correspond to data objects and hyperedges represent the communication requirements. The basic partitioning problem is to partition the vertices into k approximately equal sets such that the number of cut hyperedges is minimized. Most partitioners (including Zoltan-PHG) allows a more general model where both vertices and hyperedges can be assigned weights. It has been shown that the hypergraph model gives a more accurate representation of communication cost (volume) than the graph model. In particular, for sparse matrix-vector multiplication, the hypergraph model exactly represents communication volume. Sparse matrices can be partitioned either along rows or columns; in the row-net model the columns are vertices and each row corresponds to an hyperedge, while in the column-net model the roles of vertices and hyperedges are reversed.
Zoltan contains a native parallel hypergraph partitioner, called PHG
(Parallel HyperGraph partitioner). In addition, Zoltan provides
access to PaToH,
a serial hypergraph partitioner.
Note that PaToH is not part of Zoltan and should be obtained
separately from the
PaToH web site.
Zoltan-PHG is a fully parallel multilevel hypergraph partitioner. For
further technical description, see [Devine et al, 2006].
||PHG (parallel) or PaToH (serial)
||Check if input data is valid.
(Slows performance;intended for debugging.)
||Level of verbosity; 0 is silent.
||Print stats about final
||Desired number of processes in the vertex direction (for 2D internal layout)|
||Desired number of processes in the hyperedge direction (for 2D internal layout)|
|PHG_COARSENING_METHOD||The method to use in matching/coarsening; currently these are
agg - agglomerative inner product matching (a.k.a. heavy connectivity matching)
ipm - inner product matching (a.k.a. heavy connectivity matching)
c-ipm - column ipm; faster method based on ipm within processor columns
a-ipm - alternate between fast method (l-ipm ) and ipm
l-ipm - local ipm on each processor. Fastest option but often gives poor quality.
h-ipm - hybrid ipm that uses partial c-ipm followed by ipm on each level
||Number of vertices at which to stop coarsening.
||Ordering of vertices in greedy
0 - random
1 - natural order (as given by the query functions)
2 - increasing vertex weights
3 - increasing vertex degree
4 - increasing vertex degree, weighted by pins
||Scale edge weights by some
function of size of the hyperedges:
0 - no scaling
1 - scale by 1/(size-1) [absorption scaling]
2 - scale by 2/((size*size-1)) [clique scaling]
||Variations in "inner product"
similarity metric (for matching):
0 - Euclidean inner product: <x,y>
1 - cosine similarity: <x,y>/(|x|*|y|)
2 - <x,y>/(|x|^2 * |y|^2)
3 - scale by sqrt of vertex weights
4 - scale by vertex weights
|PHG_COARSEPARTITION_METHOD||Method to partition the coarsest (smallest) hypergraph;
typically done in serial:
random - random
linear - linear (natural) order
greedy - greedy method based on minimizing cuts
auto - automatically select from the above methods (in parallel, the processes will do different methods)
fm - two-way approximate FM
none - no refinement
|PHG_REFINEMENT_LOOP_LIMIT||Loop limit in FM refinement. Higher number means more
||Maximum number of negative moves allowed in FM.
||Controls how the balance tolerance is adjusted at
each level of bisection.
||Randomize layout of vertices and
hyperedges in internal parallel 2D layout? (0/1)
|PHG_EDGE_WEIGHT_OPERATION||Operation to be applied to edge
weights supplied by different processes for the same hyperedge:
add - the hyperedge weight will be the sum of the supplied weights
max - the hyperedge weight will be the maximum of the supplied weights
error - if the hyperedge weights are not equal, Zoltan will flag an error, otherwise the hyperedge weight will be the value returned by the processes
||Ignore hyperedges greater than this fraction times
number of vertices.
||Memory allocation for PaToH; see
the PaToH manual for details.
||Memory allocation for PaToH; see the PaToH manual for details.|
|HYPERGRAPH_PACKAGE = PHG
|Required Query Functions:|
|ZOLTAN_OBJ_LIST_FN or ZOLTAN_FIRST_OBJ_FN/ZOLTAN_NEXT_OBJ_FN pair|
|Optional Query Functions:|
It is possible to provide the graph query functions instead of the hypergraph queries, though this is not recommended. If only graph query functions are registered, Zoltan will automatically create a hypergraph from the graph, but some information (specifically, edge weights) will be lost.