The cluster version of tNavigator was tested on the supercomputer “Lomonosov” at Moscow State University. Each cluster node consisted of two four-core Intel® Xeon 5570 Nehalem processors. Up to 512 nodes were used during the calculation, which sums to 4096 cores. A real field three-phase model with 21.8 million active grid blocks was chosen as the study object. After numerical experiments and tuning, the model calculation time was reduced from 2.5 weeks down to 19 minutes. The resulting speed-up coefficient compared to one calculation core is equal to 1328, and this, as far as we know, can be called a world record for commercial solutions in this field. As in all previous tests, we couldn’t reach parallel algorithm saturation. When the number of cores was increased from 2048 to 4096, the calculation speed grew 1.4.
Speed-up for a 21.8 million active grid block model on a cluster
(with respect to one core calculation time).