Hi all, I have been testing to use Spark Graphx to do large sparse matrix multiplication for 3D image reconstruction. I used pregal API to forward and back project the images based on a graph respresentation of a large sparse matrix. I was wondering how one can optimize the Pregal operation with respect to the number of partitions that I used for graph parallelization and how I cache the intermediate variables with respect to the system that I will run on (a local multicore server vs. a super computer cluster)
Thanks a lot for your time and help in advance! Cheers, Clare