大规模油藏数值模拟软件并行计算技术及在Beowulf系统上的应用进展

THE PARALLEL COMPUTING TECHNIQUE OF LARGE SCALE PETROLEUM RESERVOIR SIMULATION SOFTWARE AND IT'S ADVANCE IN INDUSTRIAL TEST CASES ON BEOWULF SYSTEMS

本文主要介绍了大规模油藏数值模拟并行计算技术在国内的研究进展,提供了精细油藏模拟在国产Beowulf系统上的计算实例和应用效果,给出了百万网格点规模的油藏应用算例在不同处理器规模下的数值模拟计算结果与性能分析,并实现了一个针对海量数据可视化的三维图、二维图、表格显示的后处理显示系统．

This paper reports some advances of a petroleum reservoir simulation software in large scale parallel computing, an effective and efficient solver of nonlinear equations, preconditioning, and industrial applications on Beowulf systems. Three industry cases are used to test performance of this parallel reservoir simulator on several parallel hardware platforms, and the analysis and evaluation of the reported running results are given. The paper also introduces the requirements of least computation costs of finery reservoir simulations, the requirements of parallel computer hard/soft wares of a reservoir simulation system, computation methods and preconditioning of the nonlinear problem solver of reservoir simulation, the influences of parallel computation performance evaluation of the computer hardware infrastructure, and shows that this parallel code behaves satisfactory scalability and computational speedup without sacrificing the resolution of solving problems. Numerical results display that the total computing capability of solver PRIS on a Beowulf system speeds up to 1600 comparing with the results of 1998, the total computing capability speeds up at least 10 comparing with DaQing imported simulation system in 2000, and the parallel scalability may reach to 128-256 CPUs. The relative visualization system of PRIS is given out at the end of the paper.