I ran a simulation on a 3D space, which is bounds on each dimension on -1000..1000.
I execute it on 8 different node, each node processing 1/8 of the space.
For example, the first node process x=-1000..0, y=-1000..0 z=-1000..0, the second one process : x=-1000..0, y=-1000..0 z=0..1000
The output is an dataset in the same dimension, and so i got 8 "block" of dataset.
I would like to recreate the complete dataset.
If the memory in the datasets were contiguous, it would not have been possible, but as long as all these datasets are chunked, i see no problem into merging chunk into a datasets.
I've looked through family and multi driver, and i've not find any way to accomplish that.
Am i asking the impossible?