Faculty of Computing and Information Science >
Center for Advance Computing >
Cornell Theory Center Technical Reports >
Please use this identifier to cite or link to this item:
|Title: ||Parallel Finite Element Analysis of Biomechanical Structures on the Ncube 6400|
|Authors: ||Chinchalkar, Shirish|
Coleman, Thomas F.
|Keywords: ||distributed memory multiprocessors|
|Issue Date: ||Aug-1991|
|Publisher: ||Cornell University|
|Abstract: ||This paper presents parallel 3-D finite element analysis for
distributed memory multiprocessors. Traditionally, finite element
analysis has been performed on sequential computers. Current research
in high performance finite element analysis shows considerable promise
for fast, efficient implementation on MIMD and SIMD computers.
This paper demonstrates the use of a standard, banded Cholesky
method for solving the finite element system of equations. The
uniformity of the underlying data distribution ensures high performance
due to load balance. Moreover, since a distributed banded Cholesky
algorithm is likely to a part of a standard parallel numerical library,
it reduces the burden on the applications programmer, making this
method simpler to implement than the substructuring method. Since a
parallel solver requires the rows of the coefficient matrix to be
distributed in a wrap fashion, it might appear that the assembly of
the element stiffness matrices would not be efficient. However, as
shown in this paper, the calculation of element stiffness matrices,
assembly and the calculation of Gauss-point stresses can be done
efficiently in parallel without any inter-processor communication.
In fact, once nodal coordinates and element connectivity is made
available to all processors, message passing is required only during
the factorization and solution stages.
The next few sections describe how parallelism was exploited during
the assembly, solution and stress recovery strages of the finite
element analysis. The parallel program developed was tested on large
3-D finite element problems arising from biomechanical structural
systems, on an Ncube 6400. High performance Basic Linear Algebra
Subprograms (BLAS) were used to improve the execution speed.|
|Appears in Collections:||Cornell Theory Center Technical Reports|
Items in eCommons are protected by copyright, with all rights reserved, unless otherwise indicated.