Parallelizing a reservoir simulator using MPI

The Message Passing Interface (MPI) was developed to address the issue of portability of parallel codes to different parallel computers. It has been implemented on Intel iPSC/860, IBM SP1/SP2, clusters of workstations and many other architectures, providing the possibility of running the same parallel code on all of these computing platforms without changing the source code. Since MPI is implemented on top of the customized communication subroutines provided by the specific vendor for a specific machine, we are concerned about the influence of MPI on the performance of parallel codes. In this paper, a reservoir simulator was implemented on iPSC/860 using both the NX communication library provided by Intel and the MPI. Extensive numerical experiments have been conducted to compare the performance and scalability of the code using MPI with that using the NX communication library. It is found that the code using MPI has maintained the same performance and scalability as the code using the customized communication library