Distributing and collecting

Distributing a dataset is rather trivial. You simply have the master thread read the dataset locally and then use the distribute dataset routine to partition the dataset to the various threads. Collecting operates in a similar manner. The various threads read the data locally and then use a collect routine to combine the various subsections. The library also offers the ability to compress dataset along an axis. An example of this would be migration, where each thread will have its own image and the master thread wants the sum of the various portions.