Reverse-time migration in parallel: A tutorial

[kwinkunks]

Reverse-time migration in parallel: A tutorial

J. Robert Fricke

https://doi.org/10.1190/1.1442553

A reverse-time migration is implemented on a fine-grain or massively parallel computer. With fine-grain architectures many processors are distributed throughout the memory space and can operate on data “in place.” In addition, via a general communication system, any processor can access data from anywhere in the entire memory-processor space. Thus, operations on both local and global data elements are possible. These capabilities are controlled by parallel language constructs which allow parallel variable declaration, parallel arithmetic operation, and parallel random memory access. Reverse-time migration was programmed on a fine-grain machine with these hardware and software features. The reverse-time migration process had a speed improvement of two orders of magnitude relative to a state-of-the-art serial machine. At least another order of magnitude performance can be achieved with currently available floating-point processors. Similar increases in performance are expected for other seismic processes such as velocity estimation, data interpolation, 2-D filtering, and others.