Using ARMCI directly
Need 1-sided communication w/o the GA infrastructure
programmer manages distributed data structures
Good performance
- e.g., 5uS latency, 320MB/s bandwidth on the NERSC Cray T3E
Simple programming model (unlike MPI-2 1-sided)
Requires a message passing library to run (MPI,PVM,TCGMSG)
Memory allocation via ARMCI_Malloc