Challenges in High Performance Computing for Reservoir Simulation (SPE 152414)

High resolution reservoir modeling is necessary to analyze complex flow phenomena in the reservoir. As more powerful computing platforms are becoming available, simulation engineers are building larger high resolution reservoir models to study giant fields. Complexities in simulation algorithms and memory contentions are challenges to efficient use of emerging computing platforms. As new tools are available and performance of hardware and software improves, it is important to reevaluate and revise implementation details of the simulator to maintain a high level of scalability on the large number of processors. There is enormous potential in emerging technologies, such as graphic processing units (GPUs), but currently, the lack of development tools restricts their adaptation in high performance computing. Researchers in many areas of computational fluid dynamics have been able to achieve a very high level of computational performance. Such a high level is yet to be achieved in reservoir simulation. In this study, we review computational difficulties in high performance computing related to reservoir simulation and examine how improvements can be made by use of emerging technologies in this class of problems. We discuss our efforts towards improving communication and input/output (I/O) algorithms in our reservoir simulator. We also evaluate high speed interconnections, various communication libraries, etc., for parallel computations and examine how simulation performance can be improved on the latest multi-core processors. Benchmark results of various computational and I/O kernels and a summary of actual simulation results will be reviewed to illustrate current challenges and the near term outlook of high performance reservoir simulation studies.