From: George Karypis <karypis@cs.umn.edu>
Newsgroups: comp.parallel.mpi
Subject: [ANNOUNCEMENT] ParMETIS 2.0: A Parallel Graph Partitioning and
    Sparse Matrix Ordering Library
Date: Fri, 25 Sep 1998 07:41:31 -0500
Organization: University of Minnesota, Twin Cities Campus
Message-Id: <360B8F7B.F1E00E4A@cs.umn.edu>
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit


ParMETIS 2.0: A Parallel Graph Partitioning and Sparse Matrix Ordering
Library
------------------------------------------------------------------------------

URL: http://www.cs.umn.edu/~metis
URL: http://www.cs.umn.edu/~karypis/metis


We would like to announce the release of version 2.0 of the ParMETIS
library.
ParMETIS is an MPI-based parallel library that implements a variety of
algorithms for partitioning unstructured graphs and for computing
fill-reducing orderings for sparse matrices. ParMETIS is particularly
suited
for parallel numerical simulations involving large unstructured meshes.
For these computations, ParMETIS dramatically reduces the time spent in
communication by decomposing the mesh in a way that balances the load
and
minimizes the number of interface elements.

ParMETIS's algorithms are based on the multilevel partitioning and fill
reducing ordering algorithms that are implemented in the widely used
serial
package METIS. ParMETIS extends the functionality provided by METIS by
including routines that are especially suited for parallel computations
and
large scale numerical simulations.

ParMETIS provides the following four major functions:
  - Partition an unstructured graph.
  - Improve the quality of an existing partition.
  - Repartition a graph that corresponds to an adaptively refined mesh.
  - Compute a fill-reducing ordering for sparse direct factorization.


Here is a list of the major changes in version 2.0
  - Improved support for repartitioning graphs corresponding to
adaptively
    refined meshes.
    * Two new adaptive repartitioning routines have been added that are
based
      on the remapping paradigm.
    * The directed diffusion algorithm has been improved and uses a
newly
      developed wavefront formulation.
  - The number of partitions have been de-coupled from the number of
processors.
    you can now use the parallel partitioning algorithms to compute a
k-way
    partitioning independent of the number of processors that are used.
This
    is particularly useful for parallel computers consisting of clusters
of
    SMPs.
  - The names and calling sequences of all the routines have changed to
make it
    easier to call the various routines from Fortran.
  - The partitioning and ordering algorithms in ParMETIS now utilize
various
    portions of the serial METIS library. As a result, the quality of
the
    partitionings and orderings have been improved.


Obtaining ParMETIS
------------------

ParMETIS is distributed freely. Information on how to download the
source
code is available on WWW at

  URL: http://www.cs.umn.edu/~metis
or
  URL: http://www.cs.umn.edu/~karypis/metis


ParMETIS has been written by George Karypis, at the Computer Science
Department of the University of Minnesota. If you have any questions or
problems obtaining ParMETIS, send email to metis@cs.umn.edu.