<!-------- @HEADER
!
! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
!
! Zoltan Toolkit for Load-balancing, Partitioning, Ordering and Coloring
! Copyright 2012 Sandia Corporation
!
! Under the terms of Contract DE-AC04-94AL85000 with Sandia Corporation,
! the U.S. Government retains certain rights in this software.
!
! Redistribution and use in source and binary forms, with or without
! modification, are permitted provided that the following conditions are
! met:
!
! 1. Redistributions of source code must retain the above copyright
! notice, this list of conditions and the following disclaimer.
!
! 2. Redistributions in binary form must reproduce the above copyright
! notice, this list of conditions and the following disclaimer in the
! documentation and/or other materials provided with the distribution.
!
! 3. Neither the name of the Corporation nor the names of the
! contributors may be used to endorse or promote products derived from
! this software without specific prior written permission.
!
! THIS SOFTWARE IS PROVIDED BY SANDIA CORPORATION "AS IS" AND ANY
! EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
! IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
! PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL SANDIA CORPORATION OR THE
! CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
! EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
! PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
! PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
! LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
! NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
! SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
!
! Questions? Contact Karen Devine kddevin@sandia.gov
! Erik Boman egboman@sandia.gov
!
! !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
!
! @HEADER
------->
<!DOCTYPE html PUBLIC "-//w3c//dtd html 4.0 transitional//en">
<html>
<head>
<meta http-equiv="Content-Type"
content="text/html; charset=iso-8859-1">
<meta name="GENERATOR"
content="Mozilla/4.76 [en] (X11; U; Linux 2.4.2-2smp i686) [Netscape]">
<meta name="sandia.approved" content="SAND99-1377">
<meta name="author" content="karen devine, kddevin@sandia.gov">
<title>Zoltan Developer's Guide: Hypergraph Partitioning</title>
</head>
<body bgcolor="#ffffff">
<div align="right"><b><i><a href="dev.html">Zoltan Developer's Guide</a>
| <a href="dev_reftree.html">Next</a> | <a
href="dev_parmetis.html">Previous</a></i></b></div>
<h2>
<a name="Hypergraph Partitioning"></a>Appendix: Hypergraph Partitioning</h2>
Hypergraph partitioning is a useful partitioning and
load balancing method when connectivity data is available. It can be
viewed as a more sophisticated alternative to
the traditional graph partitioning.
<p>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 <i>k</i>
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
<strong>exactly</strong> 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. </p>
<p>Zoltan contains a native parallel hypergraph partitioner, called PHG
(Parallel HyperGraph partitioner). In addition, Zoltan provides
access to <a href="https://bmi.osu.edu/%7Eumit/software.htm">PaToH</a>,
a serial hypergraph partitioner.
Note that PaToH is not part of Zoltan and should be obtained
separately from the <a href="https://bmi.osu.edu/%7Eumit/software.htm">
PaToH web site</a>.
Zoltan-PHG is a fully parallel multilevel hypergraph partitioner. For
further technical description, see <a
href="ug_refs.html#hypergraph-ipdps06">[Devine et al, 2006]</a>.<br>
</p>
<h4>Algorithm:</h4>
The algorithm used is multilevel hypergraph partitioning. For
coarsening, several versions of inner product (heavy connectivity)
matching are available.
The refinement is based on Fiduccia-Mattheysis (FM) but in parallel it
is only an approximation.
<h4>Parallel implementation:</h4>
A novel feature of our parallel implementation is that we use a 2D
distribution of the hypergraph. That is, each processor owns partial
data about some vertices and some hyperedges. The processors are
logically organized in a 2D grid as well. Most communication is limited
to either a processor row or column. This design should allow for
good scalability on large number of processors.<br>
<h4>Data structures:</h4>
The hypergraph is the most important data structure. This is stored as
a compressed sparse matrix. Note that in parallel, each processor owns
a local part of the global hypergraph
(a submatrix of the whole matrix).
The hypergraph data type is <i>struct HGraph</i>, and contains
information like number of vertices, hyperedges, pins, compressed
storage of all pins, optional vertex and edge weights, pointers
to relevant communicators, and more. One cryptic notation needs an
explanation: The arrays <i>hindex, hvertex</i> are used to
look up vertex info given a hyperedge, and <i>vindex, vedge</i> are
used to look up hyperedge info given a vertex. Essentially,
we store the hypergraph as a sparse matrix in both CSR and CSC formats.
This doubles the memory cost but gives better performance.
The data on each processor is stored using local indexing, starting at zero.
In order to get the global vertex or edge number, use the macros
<i>VTX_LNO_TO_GNO</i> and <i>EDGE_LNO_TO_GNO</i>. These macros will
look up the correct offsets (using the dist_x and dist_y arrays).
Note that <i>phg->nVtx</i> is always the local number of vertices,
which may be zero on some processors.
<h4>Parameters:</h4>
In the User's Guide, only the most essential parameters have been
documented. There are several other parameters, intended for developers
and perhaps expert "power" users. We give a more complete list of all
parameters below. Note that these parameters <span
style="font-style: italic;">may change in future versions!<br>
</span>
For a precise list of parameters in a particular version of Zoltan, look at the source code (phg.c).
<table nosave="" width="100%">
<tbody>
<tr>
<td valign="top"><b>Method String:</b></td>
<td><b>HYPERGRAPH</b></td>
</tr>
<tr>
<td><b>Parameters:</b></td>
<td><br>
</td>
</tr>
<tr>
<td style="vertical-align: top;"> <span
style="font-style: italic;">HYPERGRAPH_PACKAGE</span><br>
</td>
<td style="vertical-align: top;">PHG (parallel) or PaToH (serial)<br>
</td>
</tr>
<tr>
<td style="vertical-align: top;"> <span
style="font-style: italic;">CHECK_HYPERGRAPH</span><br>
</td>
<td style="vertical-align: top;">Check if input data is valid.
(Slows performance;intended for debugging.)<br>
</td>
</tr>
<tr>
<td style="vertical-align: top;"><span style="font-style: italic;">
PHG_OUTPUT_LEVEL</span><br>
</td>
<td style="vertical-align: top;">Level of verbosity; 0 is silent.<br>
</td>
</tr>
<tr>
<td style="vertical-align: top;"> <span
style="font-style: italic;">PHG_FINAL_OUTPUT</span><br>
</td>
<td style="vertical-align: top;">Print stats about final
partition? (0/1)<br>
</td>
</tr>
<tr>
<td style="vertical-align: top;"> <span
style="font-style: italic;">PHG_NPROC_VERTEX</span><br>
</td>
<td style="vertical-align: top;">Desired number of processes in
the vertex direction (for 2D internal layout) </td>
</tr>
<tr>
<td style="vertical-align: top;"> <span
style="font-style: italic;">PHG_NPROC_HEDGE</span><br>
</td>
<td style="vertical-align: top;">Desired number of processes in
the hyperedge direction (for 2D internal layout) </td>
</tr>
<tr>
<td valign="top"><i> PHG_COARSENING_METHOD</i></td>
<td>The method to use in matching/coarsening; currently these are
available. <br>
<span style="font-style: italic;">agg</span> - agglomerative inner product
matching (a.k.a. heavy connectivity matching) <br>
<span style="font-style: italic;">ipm</span> - inner product
matching (a.k.a. heavy connectivity matching) <br>
<span style="font-style: italic;">c-ipm</span> - column
ipm; faster method based on ipm within processor columns <br>
<span style="font-style: italic;">a-ipm </span>- alternate
between fast method (l-ipm ) and ipm <br>
<span style="font-style: italic;">l-ipm </span>- local ipm
on each processor. Fastest option but often gives poor quality. <br>
<i>h-ipm - </i>hybrid ipm that uses partial c-ipm followed
by ipm on each level <br>
<i><br>
</i></td>
</tr>
<tr>
<td> <span style="font-style: italic;">PHG_COARSENING_LIMIT</span><br>
</td>
<td>Number of vertices at which to stop coarsening.<br>
</td>
</tr>
<tr>
<td style="vertical-align: top;"> <span
style="font-style: italic;">PHG_VERTEX_VISIT_ORDER</span><br>
</td>
<td style="vertical-align: top;">Ordering of vertices in greedy
matching scheme:<br>
0 - random<br>
1 - natural order (as given by the query functions)<br>
2 - increasing vertex weights<br>
3 - increasing vertex degree<br>
4 - increasing vertex degree, weighted by pins<br>
</td>
</tr>
<tr>
<td style="vertical-align: top;"> <span
style="font-style: italic;">PHG_EDGE_SCALING</span><br>
</td>
<td style="vertical-align: top;">Scale edge weights by some
function of size of the hyperedges:<br>
0 - no scaling<br>
1 - scale by 1/(size-1) [absorption scaling]<br>
2 - scale by 2/((size*size-1)) [clique scaling]<br>
</td>
</tr>
<tr>
<td style="vertical-align: top;"> <span
style="font-style: italic;">PHG_VERTEX_SCALING</span><br>
</td>
<td style="vertical-align: top;">Variations in "inner product"
similarity metric (for matching):<br>
0 - Euclidean inner product: <x,y><br>
1 - cosine similarity: <x,y>/(|x|*|y|)<br>
2 - <x,y>/(|x|^2 * |y|^2)<br>
3 - scale by sqrt of vertex weights<br>
4 - scale by vertex weights<br>
</td>
</tr>
<tr>
<td valign="top"> <i>PHG_COARSEPARTITION_METHOD</i></td>
<td>Method to partition the coarsest (smallest) hypergraph;
typically done in serial:<br>
<span style="font-style: italic;">random</span> - random<br>
<span style="font-style: italic;">linear</span> - linear
(natural) order<br>
<span style="font-style: italic;">greedy </span>- greedy method
based on minimizing cuts<br>
<span style="font-style: italic;">auto </span>- automatically
select from the above methods (in parallel, the processes will do
different methods)<br>
</td>
</tr>
<tr>
<td style="vertical-align: top;"> <span
style="font-style: italic;">PHG_REFINEMENT_METHOD</span><br>
</td>
<td style="vertical-align: top;">Refinement algorithm:<br>
<span style="font-style: italic;">fm </span>- two-way
approximate FM<br>
<span style="font-style: italic;">none</span> - no refinement<br>
</td>
</tr>
<tr>
<td> <i>PHG_REFINEMENT_LOOP_LIMIT</i></td>
<td>Loop limit in FM refinement. Higher number means more
refinement. <br>
</td>
</tr>
<tr nosave="" valign="top">
<td> <span style="font-style: italic;">PHG_REFINEMENT_MAX_NEG_MOVE</span><br>
</td>
<td nosave="">Maximum number of negative moves allowed in FM.<br>
</td>
</tr>
<tr nosave="" valign="top">
<td> <span style="font-style: italic;">PHG_BAL_TOL_ADJUSTMENT</span><br>
</td>
<td nosave="">Controls how the balance tolerance is adjusted at
each level of bisection.<br>
</td>
</tr>
<tr>
<td style="vertical-align: top;"> <span
style="font-style: italic;">PHG_RANDOMIZE_INPUT</span><br>
</td>
<td style="vertical-align: top;">Randomize layout of vertices and
hyperedges in internal parallel 2D layout? (0/1)<br>
</td>
</tr>
<tr>
<td style="vertical-align: top;"> <a
name="PHG_EDGE_WEIGHT_OPERATION"></a><span style="font-style: italic;">PHG_EDGE_WEIGHT_OPERATION</span>
</td>
<td style="vertical-align: top;">Operation to be applied to edge
weights supplied by different processes for the same hyperedge:<br>
<i>add</i> - the hyperedge weight will be the sum of the supplied
weights<br>
<i>max</i> - the hyperedge weight will be the maximum of the
supplied weights<br>
<i>error</i> - if the hyperedge weights are not equal, Zoltan
will flag an error, otherwise the hyperedge weight will be the value
returned by the processes<br>
</td>
</tr>
<tr nosave="" valign="top">
<td> <span style="font-style: italic;">EDGE_SIZE_THRESHOLD</span><br>
</td>
<td nosave="">Ignore hyperedges greater than this fraction times
number of vertices.<br>
</td>
</tr>
<tr>
<td style="vertical-align: top;"> <span
style="font-style: italic;">PATOH_ALLOC_POOL0</span><br>
</td>
<td style="vertical-align: top;">Memory allocation for PaToH; see
the PaToH manual for details.<br>
</td>
</tr>
<tr>
<td style="vertical-align: top;"> <span
style="font-style: italic;">PATOH_ALLOC_POOL1</span><br>
</td>
<td style="vertical-align: top;">Memory allocation for PaToH; see
the PaToH manual for details.</td>
</tr>
<tr>
<td valign="top"><b>Default values:</b></td>
<td><br>
</td>
</tr>
<tr>
<td><br>
</td>
<td><i>HYPERGRAPH_PACKAGE = PHG<br>
</i></td>
</tr>
<tr>
<td style="vertical-align: top;"><br>
</td>
<td style="vertical-align: top;"><span style="font-style: italic;">CHECK_HYPERGRAPH</span>
= 0<br>
</td>
</tr>
<tr>
<td><br>
</td>
<td><span style="font-style: italic;">PHG_OUTPUT_LEVEL=0</span></td>
</tr>
<tr>
<td><br>
</td>
<td><span style="font-style: italic;">PHG_FINAL_OUTPUT=0</span></td>
</tr>
<tr>
<td><br>
</td>
<td><i>PHG_REDUCTION_METHOD=ipm</i></td>
</tr>
<tr>
<td><br>
</td>
<td><span style="font-style: italic;">PHG_REDUCTION_LIMIT=100</span></td>
</tr>
<tr>
<td><br>
</td>
<td><span style="font-style: italic;">PHG_VERTEX_VISIT_ORDER=0</span></td>
</tr>
<tr>
<td><br>
</td>
<td><span style="font-style: italic;">PHG_EDGE_SCALING=0</span></td>
</tr>
<tr>
<td><br>
</td>
<td><span style="font-style: italic;">PHG_VERTEX_SCALING=0</span></td>
</tr>
<tr>
<td style="vertical-align: top;"><br>
</td>
<td style="vertical-align: top;"><i>PHG_COARSEPARTITION_METHOD=greedy</i></td>
</tr>
<tr>
<td style="vertical-align: top;"><br>
</td>
<td style="vertical-align: top;"><span style="font-style: italic;">PHG_REFINEMENT_METHOD=fm</span></td>
</tr>
<tr>
<td style="vertical-align: top;"><br>
</td>
<td style="vertical-align: top;"><i>PHG_REFINEMENT_LOOP_LIMIT=10</i></td>
</tr>
<tr>
<td style="vertical-align: top;"><br>
</td>
<td style="vertical-align: top;"><span style="font-style: italic;">PHG_REFINEMENT_MAX_NEG_MOVE=100</span></td>
</tr>
<tr>
<td style="vertical-align: top;"><br>
</td>
<td style="vertical-align: top;"><span style="font-style: italic;">PHG_BAL_TOL_ADJUSTMENT=0.7</span></td>
</tr>
<tr>
<td style="vertical-align: top;"><br>
</td>
<td style="vertical-align: top;"><span style="font-style: italic;">PHG_RANDOMIZE_INPUT=0</span></td>
</tr>
<tr>
<td><br>
</td>
<td><span style="font-style: italic;">PHG_EDGE_WEIGHT_OPERATION=max</span></td>
</tr>
<tr>
<td style="vertical-align: top;"><br>
</td>
<td style="vertical-align: top;"><span style="font-style: italic;">EDGE_SIZE_THRESHOLD=0.25</span></td>
</tr>
<tr>
<td style="vertical-align: top;"><br>
</td>
<td style="vertical-align: top;"><span style="font-style: italic;">PATOH_ALLOC_POOL0=0</span></td>
</tr>
<tr>
<td style="vertical-align: top;"><br>
</td>
<td style="vertical-align: top;"><span style="font-style: italic;">PATOH_ALLOC_POOL1=0</span></td>
</tr>
<tr>
<td valign="top"><b>Required Query Functions:</b></td>
<td><br>
</td>
</tr>
<tr>
<td><br>
</td>
<td><b><a href="../ug_html/ug_query_lb.html#ZOLTAN_NUM_OBJ_FN">ZOLTAN_NUM_OBJ_FN</a></b></td>
</tr>
<tr>
<td><br>
</td>
<td><b><a href="../ug_html/ug_query_lb.html#ZOLTAN_OBJ_LIST_FN">ZOLTAN_OBJ_LIST_FN</a></b>
or <b><a href="../ug_html/ug_query_lb.html#ZOLTAN_FIRST_OBJ_FN">ZOLTAN_FIRST_OBJ_FN</a></b>/<b><a
href="../ug_html/ug_query_lb.html#ZOLTAN_NEXT_OBJ_FN">ZOLTAN_NEXT_OBJ_FN</a></b>
pair</td>
</tr>
<tr nosave="" valign="top">
<td><br>
</td>
<td nosave=""> <b><a href="../ug_html/ug_query_lb.html#ZOLTAN_HG_SIZE_CS_FN">ZOLTAN_HG_SIZE_CS_FN</a></b>
<br>
<b><a href="../ug_html/ug_query_lb.html#ZOLTAN_HG_CS_FN">ZOLTAN_HG_CS_FN</a></b>
</td>
</tr>
<tr>
<td valign="top"><b>Optional Query Functions:</b></td>
<td><br>
</td>
</tr>
<tr>
<td><br>
</td>
<td><b><a href="../ug_html/ug_query_lb.html#ZOLTAN_HG_SIZE_EDGE_WTS_FN">ZOLTAN_HG_SIZE_EDGE_WTS_FN</a></b></td>
</tr>
<tr>
<td><br>
</td>
<td><b><a href="../ug_html/ug_query_lb.html#ZOLTAN_HG_EDGE_WTS_FN">ZOLTAN_HG_EDGE_WTS_FN</a></b></td>
</tr>
</tbody>
</table>
<p>
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. </p>
<hr width="100%">[<a href="ug.html">Table of Contents</a> | <a
href="dev_reftree.html">Next:
Refinement Tree Partitioning</a> | <a
href="dev_parmetis.html">Previous:
ParMetis</a> | <a href="https://www.sandia.gov/general/privacy-security/index.html">Privacy and Security</a>]
</body>
</html>