@device(postscript)
@libraryfile(Mathematics10)
@libraryfile(Accents)
@style(fontfamily=timesroman,fontscale=11)
@pagefooting(immediate, left "@c<Technical Report 1996>",
center "@c<Computer Science>",
right "@c<Carnegie Mellon>")
@heading(Properties of a Family of Parallel Finite Element
Simulations)
@heading(CMU-CS-96-141)
@center(@b(David R. O'Hallaron, Jonathan R. Shewchuk))
@center(December 1996)
@center(FTP: CMU-CS-96-141.ps)
@blankspace(1)
@begin(text)

This report characterizes a family of unstructured 3D finite element
simulations that are partitioned for execution on a parallel system.
The simulations, which estimate earthquake-induced ground motion in
the San Fernando Valley of Southern California, range in size from
10,000@y(M)1,000,000 nodes and are partitioned for execution on 4@y(M)128
processors.  The purpose of the report is to help researchers better
understand the properties of unstructured tetrahedral finite element
meshes and the sparse matrix vector product (SMVP) operations that are
induced from them. The report is designed to serve as a comprehensive
reference that researchers can consult for answers to the following
kinds of questions: For a tetrahedral mesh with a particular number of
nodes, how many elements and edges does it have?  What is the
distribution of node degrees in a tetrahedral mesh? What fraction of
nodes in a partitioned mesh are interface nodes? What is the
communication volume in a typical parallel SMVP?  How many messages
are there? How big are the messages?  How many nonzeros are contained
in the rows of a sparse matrices induced from tetrahedral meshes?  The
partitioned meshes described in the paper are available
electronically.

@blankspace(2line)
@begin(transparent,size=10)
@b(Keywords:@ )@c<Unstructured finite element meshes, parallel computing,
sparse matrix>
@end(transparent)
@blankspace(1line)
@end(text)
@flushright(@b[(22 pages)])