Next: Parallel calculation of the
Up: parallel implementation
Previous: parallel implementation
We have distributed the calculation of the weight matrix on a beowulf cluster using
the parallel infrastructure described in Clapp (2005).
The complete image and masks are distributed to each node. This is necessary because the random bounds requires random jumps around the image.
Different jobs are assigned different rows of the weight matrix, no communication
beween nodes is necessary while calculating the weight matrix. Upon completion,
the weights are collected into a single sparse matrix on the master node
Next: Parallel calculation of the
Up: parallel implementation
Previous: parallel implementation
Stanford Exploration Project
4/5/2006