ComEx Performance
The following figure shows the performance of the Blocking Get operation on the NERSC Edison system (--with-dmapp), which uses the Cray Aries Network.

The following figure shows the performance of the Blocking Get operation on the NERSC Hopper system (--with-dmapp), which uses the Cray Gemini Network.

The following figure shows the performance of the Blocking Get operation using the OpenFabrics (--with-ofa) device on the PNNL Institutional Computing (PIC) system, which uses Mellanox InfiniBand QDR Network.

The following figure shows the performance of the Blocking Get operation using the MPI Two-sided compatibility port (--with-mpi-ts), which uses MPI send/recv communication primitives to implement the one-sided primitives of ComEx. This port is intended to for desktops, laptops, or any system which is not natively supported by ComEx.
