Hybrid MPI-OpenMP scalable parallelization for coupled non-isothermal fluid-heat flow and elastoplastic geomechanics Conference Paper uri icon

abstract

  • 2017, Society of Petroleum Engineers We parallelize individual non-isothermal fluid-flow and geomechanics simulators separately, and then use a sequential method for coupling between flow and geomechanics. Message Passing Interface (MPI) is employed in the distributed memory system, and Open Multi-Processing (OpenMP) is used in the shared memory system. We primarily implement MPI for matrix assembly and parallel solvers, particularly using the PETSc library codes while using OpenMP for other miscellaneous subroutines to prevent significant overheads. We also study different matrix decomposition schemes for the geomechanics linear system, which not only precondition for parallel solvers but also assign computation loads to cores involved in the cluster. We take more than one million cells to investigate parallel performance. For both flow and geomechanics, the parallelization largely reduces the overall simulation execution time and obtains scalable speedups. Parallel-solver-only performance is tested as well. We find that solver performance achieves scalable speedups, where the optimum overall speedup for the parallel coupled simulator exceeds 14. Also, matrix decomposition methods have effects on parallelization in terms of execution time and solver performance, implying that the selection of matrix decomposition methods is important for high efficiency parallelization. The parallel scheme in this study can straightforwardly be applied to other sequentially coupled simulators, without significant code development. When the flow and geomechanics simulators are sequentially coupled, we find that the scalability of the parallel coupled simulator can be honored. We also find that plasticity can lead to imbalance in the parallel environment. Thus, efficient parallel schemes for plasticity are required.

published proceedings

  • Society of Petroleum Engineers - SPE Reservoir Simulation Conference 2017

author list (cited authors)

  • Guo, X., Kim, J., & Killough, J. E.