We describe experiences with parallelizing real models using an unstructured reservoir simulator on shared memory machines. Using ExxonMobil's proprietary reservoir simulator, EMpowerTM, which contains an object oriented data hierarchy which describes a fully unstructured mesh, a fully integrated well and surface facility network, and a highly flexible, user specified well management capability, we will show how parallelization is performed at a high level through several variants of data partitioning adapted to the specific algorithmic needs. Property calculations, Jacobian construction, matrix solution methods, and well management may use a variety of different data structures, partitioning, and parallel execution schemes. We discuss mechanisms for addressing problems encountered.

You can access this article if you purchase or spend a download.