The fundamental of cross-node elastic modeling is to split one-node-run job to various sub-jobs with multiple nodes, each node will also run with multi-threading configuration. Cross-node technology significantly improves the computation efficiency of large scale 3D elastic modeling. The idea is simple, along one geometry direction, for example, inline direction (Y direction) in 3D model, split samples in inline direction to multiple small parts, then submit each part to different computer nodes:
In this algorithm, since Finite Difference algorithm will update volume information at every time step, we need to assign a master node to collect data from other real computational nodes, exchange boundary information, update each block information:
The basic workflow is described as:
The efficiency of this algorithm is also depend on the speed of network. If you have infinite bandwidth, the time for exchange boundary information among each node will be minimum. If you have relatively narrow bandwidth, more computer nodes may decrease the overall efficiency.
(All technologies are integrated in SeisSpace processing software and Halliburton/Landmark own Intellectual property.)