Development of a Parallel Platform for Reservoir Simulations on Supercomputers

We introduce our work to speed reservoir simulations by using supercomputers, such as IBM Bluegene/Q and regular clusters. We have developed a parallel platform, which provides grid, data, linear solvers, preconditioners, key... [ view full abstract ]

We introduce our work to speed reservoir simulations by using supercomputers, such as IBM Bluegene/Q and regular clusters. We have developed a parallel platform, which provides grid, data, linear solvers, preconditioners, key word parsing, well modeling and visualization modules. General graph and geometry load balancing modules are provided to distribute meshes and minimize communication, including the Hilbert space-filling curve method and ParMETIS. The grid module supports finite difference methods and finite volume methods. Commonly used Krylov subspaces solvers, including GMRES, BICGSTAB, CGS and ORTHOMIN, and efficient preconditioners, including restricted additive Schwarz, ILU(k), ILUT(p, tol), block ILU(k), algebraic multi-grid and a family of Constrained Pressure Residual preconditioners, are implemented. Based on this platform, parallel reservoir simulators are being developed. At this moment, a two-phase oil-water simulator, a three-phase black oil simulator and a multiphase thermal simulator have been implemented.

Numerical experiments have been performed on IBM Bluegene/Q and other supercomputers. Simulations with billions of grid cells using thousands of CPU cores have been tested. And the scalability of our platform is linear. The numerical experiments show that our platform has good designs and it is capable of speeding simulation thousands of times.