Newsgroups: comp.parallel
From: wc06 <wc06%vogon.uucp@nac.no>
Subject: ANNOUNCE: jostle - unstructed mesh partitioning software
Organization: School of Maths, University of Greenwich, U.K.
Date: Fri, 10 Mar 1995 16:07:43 +0000
Message-ID: <3k4bv3$aqi@usenet.srv.cis.pitt.edu>

We would like to announce the release of the JOSTLE software package
for partitioning unstructured meshes for use on distributed memory
parallel computers. Executables of the software are freely available
for academic and research purposes, but interested users are required
to sign a licencing agreement.

The code is designed to partition unstructed meshes (for example,
finite element or finite volume meshes) and can also be used to
repartition existing partitions (such as those deriving from adaptive
refined meshes).

You can get a copy of the licence either via the WWW from
        http://www.gre.ac.uk/~wc06/jostle/licence.ps.gz
 
or I can email you a (gzipped uuencoded postscript) copy or I can put
a hardcopy in the post. Although the licence may seem a bit stringent,
its purpose is to prevent resale or commercial exploitation.
We strongly want to encourage the use of the software, however, and
would be keen to work on collaborations, etc. We are also interested
in supporting commercial exploitation, although under different licencing
arrangements. Send the completed licence to me and I will mail you
a copy of the package.
 
        Executables are available for various machines including Suns
(SPARC compiler), Suns (sun4 compiler), Silicon Graphics Indigo and IBM
RS6000. Please let me know which version(s) you require (you can have
more than one under the same licence).
 

You can find out more about the underlying algorithms from the technical
report:

    C. Walshaw, M. Cross, and M. Everett.
    A Parallelisable Algorithm for Optimising Unstructured Mesh Partitions.


Here is the abstract:

A new method is described for optimising graph  partitions  which
arise  in mapping unstructured mesh calculations to parallel com-
puters. The method employs a combination of iterative  techniques
to  both  evenly balance the workload and minimise the number and
volume of interprocessor communications.  It is designed to  work
efficiently  in  parallel  as well as sequentially and can be ap-
plied directly to dynamically refined meshes.  In addition,  when
combined  with  a fast direct partitioning technique (such as the
Greedy algorithm) to give an  initial  partition,  the  resulting
two-stage  process proves itself to be both a powerful and flexi-
ble solution to the static graph-partitioning problem. A cluster-
ing technique can also be employed to speed up the whole process.
Experiments, on graphs with up to a million nodes, indicate  that
the resulting code is up to an order of magnitude faster than ex-
isting state-of-the-art techniques such as  Multilevel  Recursive
Spectral  Bisection,  whilst  providing  partitions of equivalent
quality.

The paper can be obtained via WWW from the URL:
	http://www.gre.ac.uk/~wc06/papers/ps/Walshaw95.ps.gz

although if you have any difficulty in obtaining it, I'll be happy
to send you a copy, either by post or electronically.

Two other papers on the same topic and with further results are:--

	C. Walshaw, M. Cross, S. Johnson, and M. Everett.
	A Parallelisable Algorithm for Partitioning Unstructured Meshes.
	In: Proc. Irregular '94: Parallel Algorithms for Irregularly
	  Structured Problems (in press), 1994.

	C. Walshaw, M. Cross, S. Johnson, and M. Everett.
	JOSTLE: Partitioning of Unstructured Meshes for Massively Parallel
	  Machines.
	(submitted for the Proceedings, Parallel CFD'94, Kyoto), 1994.



	Chris Walshaw

