Parallelizing black oil reservoir simulation systems for SMP machines

We discuss the parallelization for SMP machines of black oil reservoir simulation programs. A reservoir simulator is a sophisticated computer program used to predict the future performance of a reservoir based on its current state and past performance. SMP architectures are composed of homogeneous microprocessors sharing access to the main memory. Due to advancements in microprocessor technology and the shared memory access, the best parallelization strategy for SMP (symmetric multiprocessing) machines may differ significantly from strategies used previously in MPP machines. We present a parallel implementation optimized for SMP machines of BOAST a three-phase, three-dimensional black oil simulator. The communication library used was LAM-MPI. We used the domain decomposition technique to parallelize the two more computing-intensive subprograms (routines) of BOAST which take about 80% of the total execution time of a typical simulation in a SMP machine. The speed-up obtained show that the chosen parallelization strategy has good load balancing and low communication overhead, achieving near-optimal performance in several simulations.