Reducing run time in seismic imaging computing

摘要

A system and computer program product for seismic imaging implements a seismic imaging algorithm utilizing Reverse Time Migration technique requiring large communication bandwidth and low latency to convert a parallel problem into one solved using massive domain partitioning. Since in RTM, both the forward and reverse wave propagation is iteratively calculated on time step increments, the method implements methods that includes partitioning memory between computation and intermediate results to optimize an RTM computation. The methods make maximum use of the memory to either eliminate storing the snapshot wavefield data to disk, or hide all or a significant portion of the disk I/O time. Furthermore, the schemes can provide the flexibility to vary a number of iterations (step size) for each snapshot to be kept in the memory. If any of the given conditions changes during the process, maximum usage of the available memory is ensured.

主权项

1. A system for managing available memory at computing devices to reduce seismic imaging computation time, said system comprising:
a parallel computer device having plurality of computing nodes interconnected through one or more communication networks, each node having one or more associated programmed processor units and an associated coupled first local memory storage device, a processor unit programmed to perform a method comprising:
partitioning a model space volume as domains over said multiple computing nodes,assigning each partitioned domain to one or more processing nodes;determining an available amount of said first local memory storage at each node;determining a sub-volume size per node corresponding to domain model processing;determining from said available amount of first memory and sub-space volume size a first amount of iterations corresponding to determined time steps at which forward wave propagation data is computed for storage;wherein during an iterative processing in a forward wavefield propagation calculation path, storing computed forward wavefield data contributions for said first amount of time steps needed to a second external memory storage device; and,in each remaining time steps, storing the computed forward wavefield data contributions in said available first memory; and,
wherein during an iterative processing in a backward wavefield propagation calculation path, using the stored computed forward wavefield data contributions from the available first memory during said first amount of time steps to compute reverse wave propagation data in said backward calculation path, andin a remaining time steps, using the stored computed forward wavefield data contributions from the second memory storage device to compute reverse wave propagation data; and,computing, using said stored forward wave propagation data and reverse wave propagation data, an image at each node, and,merging said computed images from each node to create a final image.