How could I use the configurable memory segments in GASPI? What are the use-cases for this?

The most frequent current use-case is to exploit the possibility to directly read and write from/to the GPGPU/Xeon Phi memory of other nodes. Essentially, GASPI allows the application to generate two different partitioned global address spaces for this case, with one address space across the local memory of the GPGPU/Xeon Phi and one address space across the main memory of the node. How much of this memory is reserved for this global address space (and how many memory segments are created) is left to the application programmer.