Performance considerations:
- used for highly asynchronous applications
- most important: overlap of communication and computation
- simple test program: linear speedup, due to good overlap
Memory considerations:
- needs ``special memory''
- allocated with MPI_Alloc_mem
- maximal size: smaller than MPI_GLOBMEMSIZE
- if too small:
- enlarge MPI_GLOBMEMSIZE
- play with MPI_SHMEMCNTL
Peter Junglas 6.3.2000