M. Ehtesham Hayder

Towards Building Scalable Grid Computing Environments for Reservoir Simulation

Standard high performance clusters (HPC) are being extensively used for solving computationally intensive problems in various scientific fields, one of which is the field of reservoir simulation. In our work, we expand the conventional HPC systems to run larger reservoir simulations on a heterogeneous grid of HPCs. This expansion is accomplished by developing a unique domain decomposition technique to achieve higher load balancing and computational efficiency on the grid. The decomposer works as an adaptive detector of the available hardware resources on the grid, learns about the different processor speeds and adjusts its load distribution among the CPUs. As a result, our grid implementation for reservoir simulation outperforms the conventional methods of domain decomposition, which are designed to run on homogenous set of processors in a cluster.

Poster: high performance computing reservoir simulation in the oil industry

Large reservoir simulation is computationally challenging. High resolution reservoir models provide necessary realism in heterogeneous reservoirs. Large number of simulations of such models is necessary in reducing uncertainties in reservoir characterizations and creating efficient field development plans. To meet such challenges, Saudi Aramco, the worlds leading oil producing company, developed a parallel reservoir simulator, POWERS [1], and built a computing center currently hosting computing power of over 300 teraflops [2]. Both reservoir simulator and computing center are world leaders in the oil industry. In this paper we review some of most innovative HPC works in the oil industry. We review computational characteristics of the POWERS simulator and benchmark results for computations of large reservoir simulation models throughout the past few years. We also share studies to enhance POWERS in various aspects such as GPU usage to solve critical computing blocks, MPI tuning, and reducing the eco footprint of the simulator.