<!-------- @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.approval_type" content="formal">
  <meta name="sandia.approved" content="SAND2007-4748W">
  <meta name="author" content="Zoltan PI">

  <title>Zoltan User's Guide: Graph vs. Hypergraph Partitioning</title>
</head>
<body bgcolor="#ffffff">
<div align="right"><b><i><a href="ug.html">Zoltan User's Guide</a>&nbsp; |
<a href="ug_alg_parmetis.html"> Next</a>&nbsp; |
&nbsp;<a href="ug_alg_graph.html">Previous</a>&nbsp;
</i>
</b></div>

<h2>
<a name="Graph vs Hypergraph Partitioning">
</a>Graph vs Hypergraph Partitioning</h2>

Graph partitioning has proven quite useful in scientific computing. 
<a href="ug_alg_hypergraph.html">Hypergraph partitioning</a>
is a more recent improvement
that uses a hypergraph model, which is
often a more accurate model than the graph model for scientific computing. 
(Hypergraphs contain hyperedges which connect two <i>or more</i>
vertices.)
See <a href="ug_refs.html#catalyurek99">[Catalyurek & Aykanat]</a> and
<a href="ug_refs.html#hendrickson-kolda">[Hendrickson & Kolda]</a> 
for further details. 

You do not need to understand the underlying models
to use graph or hypergraph partitioning for load-balancing
in Zoltan. The basic trade-offs are:
<ul>
<li> Hypergraph partitioning usually produces partitions (assignments)
of higher quality than graph partitioning, which may reduce
communication time in parallel applications (up to 30-40%
reduction has been reported). However, hypergraph
partitioning takes longer time to compute. </li>
<li> The graph model is restricted to symmetric data dependencies. 
If you have a non-symmetric problem, we recommend hypergraph 
partitioning. </li>
</ul>

<h3>Migrating from ParMetis to PHG in Zoltan</h3>
If you already use Zoltan for graph partitioning (via ParMetis),
there are three ways to switch to the Zoltan-PHG hypergraph partitioner: 

<ol>
<li>The quick and easy way: Just change the <a href="ug_alg.html#LB_METHOD">
LB_METHOD</a> to "Hypergraph".
Zoltan will then use the graph query functions (presumably
already implemented) to construct a hypergraph model, which
is similar to but not equivalent to the graph.
<li>The proper way: Change the LB_METHOD, but also implement and
register the <a href="ug_query_lb.html#ZOLTAN_HG_SIZE_CS_FN">
hypergraph query functions</a>
required by Zoltan. These may give a more accurate representation
of data dependencies (and communication requirements) for your
application.
<li>If you really want graph (not hypergraph) partitioning: 
Just change the <a href="ug_alg.html#LB_METHOD">
LB_METHOD</a> to "Graph".
Zoltan will then use PHG as a graph partitioner, which is slower than
ParMetis but often produces better partitions (lower cuts).
</ol>

Technical note: A hypergraph is constructed from the graph as follows:
The vertices are the same in the hypergraph as in the graph. For each vertex
<i>v</i>, create a hyperedge that consists of all neighbors in the graph 
and <i>v </i> itself.

<hr width="100%">[<a href="ug.html">Table of Contents</a>&nbsp; | 
<a href="ug_alg_parmetis.html"> Next:&nbsp; ParMETIS</a>&nbsp; |
&nbsp; <a href="ug_alg_graph.html">Previous:&nbsp;
Graph Partitioning</a>&nbsp; |&nbsp; <a href="https://www.sandia.gov/general/privacy-security/index.html">Privacy and Security</a>]
</body>
</html>