A RING-BASED PARALLEL OIL RESERVOIR SIMULATOR∗

Abstract. We develop and implement a ring-based parallel 3-D oil-phase homogeneous isotropic reservoir simulator and study its performance in terms of speedup as a function of problem size. The ring-based approach is shown to result in significant improvement in speedup as the problem size increases. This improvement stems from the reduction in communication costs inherent in a ring-based approach. The simulator employs a parallel conjugate gradient (CG) algorithm that we develop for solving the associated system of linear equations. The parallelization uses an MPI programming model. Previously proposed parallel oil reservoir simulators focus on data parallelism and load balancing and gives less attention to the communication cost. Performance analysis is given showing that the parallel algorithm results in a speedup of more than 42 times compared to a sequential simulator for a large simulation problem. This major improvement occurs for larger problem sizes, since the communication savings become significant. We compare our results to the implementation of the parallel oil reservoir simulator using the Portable Extensible Toolkit for Scientific Computation (PETSc). Oil reservoir simulators are used for forecasting reservoir potential before costly drilling, and are essential for improving oil recovery from existing fields, helping to maximize oil production. The speedup gained through the technique presented here can result in major savings of engineering time and more accurate reservoir management, and in turn higher oil production. Existing simulators suffer from limited performance due to the huge numerical operations involved. To cope with the issue, engineers usually reduce the size of the simulation model to get results in an acceptable timeframe, sacrificing accuracy of the predictions. This article describes the proposed ring-based algorithm for parallelization and development of a 3-