Parallel Reservoir Simulation on Shared and Distributed Memory System

Abstract

Abstract Computational efficiency is a major subject in large-scale reservoir simulations. Parallel simulation has gained popularity in recent years. Complex geological environments and production management demand new mathematics and science in various computational environments. This paper presented a parallel simulation algorithm and data processing developed for shared and distributed memory systems. The preconditioned Krylov subspace method with hybrid preconditioner based on domain decomposition was constructed to solve the large-scale linear system resulted from black-oil simulator techniques. Based on a fully implicit, three-dimensional, three-phase black-oil reservoir simulator, this project focused on the numerical intensive linear solver. The parallel algorithm is scalable and applicable to both shared and distributed memory systems. Numerical tests were performed on both multi-CPU workstations and PC clusters. A message passing interface (MPI) library was adopted for all message passing functionality in both shared and distributed memory environments. Input/output parallelism and localism are presented in these hardware setups. Numerical results from benchmark models and field application examples are presented for both the multi-CPU SGI Power Challenger and a Linux based Intel PC cluster. Results show that the speedup factor is significant. The simulation performance increases with the number of processors and only decreases due to the limitations introduced by message passing activities.