José Roberto P. Rodrigues
Reservoir simulation is a crucial component of almost every decision in the oil production industry. At the core of these reservoir simulators is the solution of very large systems of linear equations, which is typically the step requiring the highest amount of computer resources. After briefly reviewing the techniques used to model petroleum production and how they lead to the need of solving such large linear systems, this talk will discuss the iterative methods most commonly employed, highlighting specific features which are required by some peculiar aspects of those systems. The impact of modern computer architecture on the design of the linear solvers will be emphasized. Key aspects taken into account in a dedicated solver for petroleum applications currently under development will be shown, as well as results obtained with matrices from real simulations, focusing on performance and scalability.