Computer models of oil reservoirs are becoming more and more complex in order to represent the heterogeneity of rock structures more realistically. It is, however, necessary to restrict the resolution of models and the number of components in compositional models because of the limits of computer CPU time and memory. This leads to uncertainties in the simulation due to the averaging techniques used in petrophysics and in the modelling of surface processes.

Parallel computing is now an attractive solution to overcome these drawbacks by using the best of the hardware and software technology.

In this paper, we look at the efficiency of the parallel versions of Eclipse 100 and 300. We evaluate their speed, memory requirements and scalability using black oil and compositional models with up to 16 components on an SGI Origin 2000 with 16 processors. These are based on real fields and range from 100 000 to 4 500 000 cells with cells as small as 20×20×0.5 meters.

We examine the problems of deciding on the decomposition of domains, and the choice of message passing systems. We also discuss the optimal way of managing the job scheduling. We show the difficulties of pre and post processing large data sets and visualizing the results of simulations. The analysis of results of these test cases allowed us to quantify the benefits of a finer geological and fluid description. With the parallel option, we can now envisage the possibility of running weekend jobs overnight and overnight jobs during normal working hours.

We show that the jump of parallel computing from research to industrial applications provides real benefits to engineers.

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