AuthorsA. Thune and X. Cai
TitleBalancing the numerical and parallel performance for reservoir simulations
AfilliationScientific Computing
Project(s)Department of High Performance Computing
Publication TypeTalks, contributed
Year of Publication2019
Location of TalkSIAM Conference on Computational Science and Engineering (CSE19), Spokane, Washington, USA

The overall performance of a PDE-based simulator depends on two factors: the algorithmic efficiency of the numerical scheme chosen and the parallel efficiency of the software implementation. Since aspects from the two factors may influence each other's performance, a suitable balance between the two is important. The focus of this talk is on the OPM framework of oil reservoir simulation, for which the computational core is to solve the black-oil model: a coupled system of nonlinear PDEs. Due to large variations in the geological properties of a reservoir, the sparse matrix that arises from discretizing the coupled PDEs exhibits a strong heterogeneity in its nonzero values. These reflect the strength of coupling between the degrees of freedom. It is thus necessary to consider this heterogeneity in the unstructured mesh partitioning process, typically translated to partitioning a graph with weighted edges. Particularly, we study the impact of different strategies of edge weighting on both the numerical and parallel performance. The ordering of the degrees of freedom, which also affects both sides, is studied in addition. Our purpose is to shed some light on a suitable mesh partitioning and ordering methodolgy, which is also relevant beyond the context of reservoir simulation. The issue of how to allow users of OPM to inject such flexibility into the existing software framework is also discussed.

Citation Key26986