Requirement for Consistency in the Parallel Implementation of Explicit FEA Crash Simulation Programs

Failure to provide consistent results in the parallel implementation of a crash simulation can render the simulation code unusable as a design heuristic. This paper describes a parallel implementation of a crash simulation package, FCRASH, which was designed from inception for parallel implementation. An example will be given which illustrates the variation a CAE crash analyst may encounter if the problem of parallel consistency is ignored. Techniques used in FCRASH to deliver consistent results in a parallel environment on a CRAY T90 parallel supercomputer will be discussed. This work has resulted in a robust crash simulation code that delivers consistent results in parallel environments on a variety of shared memory processors. Finally, problems remaining for delivering the same level of robustness in a distributed memory message passing version of an explicit crash simulation program will be examined.