Parallel Implementation of a Commonly Used Internal Combustion Engine Code

摘要

KIVA-3, a widely used computational fluid dynamics code for internal combustion engine simulations, was re-structured for parallel execution via one-dimensional domain decomposition. Communication between processors at the subdomain boundaries is carried out with MPI message-passing library which enabled significant flexibility in porting the distributed-memory implementation of KIVA-3 to various architectures ranging from SGI Origin 2000 to Cray T3E and recently a Beowulf cluster. Due to the complex features incorporated into KIVA-3, the paralleliza-tion was done in stages. This paper presents results from the first stage where the speedup and parallel efficiency of the parallel version of KIVA-3 are elucidated.