Reservoir Simulations in Clusters of PCS

Abstract

In recent years, clusters of PCs has becoming an economical and attractive alternative to the use of traditional high performance computers, since these computers can provide a price/performance ratio up to ten times fewer then current proprietary parallel computers, making possible to small and medium officers, centers and scientific institutions, looking up to this technology for the solution their problems. This work describes the application of clusters of PCs for reservoir simulations, showing some results obtained with a parallel simulator developed for distributed memory computers. In the simulator, a two-phase Black-Oil model was implemented, and the partial differential equations governing the problem are discretized using finite differences in a 3D block centered grid, using a Cartesian coordinate system. IMPES and fully implicit procedures can be used as numerical solution, and solvers from software PETSc, witch employ MPI for communication between processors, handles the resulting systems of nonlinear equations, by means of the Inexact-Newton method using several parallel Krylov subspaces methods and preconditioners. The performance of the simulator running in a cluster of PCs is analyzed for solution of some reservoir problems. Speedup and efficiency measurements show de applicability of these computers in the solution of large-scale problems in reservoir simulation and the parallel programming model used in the parallel simulator.