A large amount of research in domain decomposition methods for parallel computing has been carried out in the last twenty years. This paper presents an automatic domain decomposition framework for parallel reservoir simulation. It emphasizes practical issues and implementation rather than theoretical complexity.

At least three main factors affect the parallel simulator performance when decomposing the domain for the given total number of sub-grids: 1) the active cell distribution in the reservoir, 2) reservoir geometry, and 3) fault locations. This discussion primarily focuses on the active cell distribution to improve the performance of the parallel reservoir simulator by load balancing.

Four domain decomposition algorithms are implemented before determining the best decomposition for the given total number of sub-domains. These four algorithms applied to all factors of the total number of sub-domains. The ideal decomposition for active cells and the offset size between the ideal number and real number from each scenario are then calculated. The smallest offset size determines the best decomposition.

Several examples, such as a synthetic model, a GOM (Gulf of Mexico) model with over 550,000+ grid blocks and a synthetic African reservoir compositional model, illustrate the improved parallel simulator performance resulting from the automatic decomposition framework. For both cases, the reservoir models were decomposed by different algorithms, and the CPU time for each algorithm was summarized. The automatic decomposition framework provides an effective way to improve the parallel simulator performance.

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