Finite element ground‐water models implemented on vector computers

Abstract

SUMMARYFinite element models, optimized for running on conventional serial computers, are not suitable to make use of the potentional high performance of today's vector or parallel computers. Also, automatic vectorization by the compiler or manual vectorization at the local level do not by far lead to the required and expected computational speed.A global change of the overall program logic or a complete redesign becomes necessary, i.e. a completely new generation of program systems has to be created, considering the new hardware characteristics and abilities.A new computer-independent programming technique for finite element problems to be implemented on vector computers is proposed and the test results of scalar and vectorized program structures on a conventional serial and on a non-conventional pipeline computer are discussed.

Supercomputers and their use in modeling subsurface solute transport

“…"Standard" FORTRAN 77 is not well suited to vector or parallel machines, and manufacturers have introduced extensions to allow machine features to be exploited. For example, Pelka and Peters [1986] demonstrate the usefulness of the Cray's GATHER/SCA•R feature for vectorizing finite element solutions of the groundwater flow equation. The research of Sudicky and coworkers cited above and that of Meyer et al [1989] demonstrate the very substantial speedups that can be gained in a wide class of transport problems.…”

“…Simply transferring codes from serial to vector or parallel machines may not result in very impressive speedups. This point is demonstrated clearly by Pelka and Peters [1986], who compare computation rates for scalar and vectorized finite element solutions of a groundwater flow model. Particularly for parallel machines, the algorithms used may turn out to be not very efficient.…”