A Hybrid Approach to Parallel Multiscale Reservoir Simulator

In the wake of advancements in multi-core/multi-processor NUMA systems and associated hardware architectures, there is a need and an opportunity for software to be enhanced in lock-step with these hardware developments, in order to reap performance gains. However, Giga cell reservoir models still may require days of simulation on available hardware. Recently in reservoir simulation various multiscale methods have been developed in order to create faster computation algorithm. The key idea of these methods is to solve the problem on a coarse level and then prolong it to fine level using a set of basis functions. These prolongation operators map between fine grid geological properties of reservoir model and the coarse grid that is used for simulation. One of these methods has been implemented in a commercially available simulator. The next natural step to further speed-up simulation is to use high performance technologies for parallelization. We illustrate how this was accomplished by reusing and extending the existing parallel pattern for distributed, MPI, and by adding shared, OpenMP, memory techniques. Finally results are presented for large highly heterogeneous real-field models showing performance and scalability of the new industrial strength multiscale reservoir simulator running on a multi-node cluster.