The general purpose finite element (EE) system PERMAS(1) has been exte
nded to support shared and distributed parallel computer architectures
as well as workstation clusters. The methods used to parallelize this
large application software package are of high generality and have th
e capability to parallelize all mathematical operations in a FE analys
is-not only the solver. Utilizing the existing hyper-matrix data struc
ture for large, sparsely populated matrices, a programming tool called
PTM was introduced that automatically parallelizes block matrix opera
tions on-the-fly. PTM totally hides parallelization from higher order
algorithms, thus giving the physically oriented expert a virtually seq
uential programming environment. An operation graph of sub-matrix oper
ations is asynchronously built and executed. A clustering algorithm di
stributes the work, performing a dynamic load balancing and exploiting
data locality. Furthermore, a distributed data management system allo
ws free data access from each node. The generality of the approach is
demonstrated by some benchmark examples dealing with different types o
f FE analyses. (C) 1998 Elsevier Science Ltd. All rights reserved.