mpc_send()-mpc_recv()

mpc_send() is an asynchronous non-blocking send and mpc_recv() is a non-blocking receive. MPL provides polling function to check the completion of the send and receive operations.

We used mpc_bsend()-mpc_brecv() to do our MPL bandwidth and latency measurements.

PVMe can be used exactly as PVM 3.2 (that is, without pvm_psend() or pvm_precv()). It can also be used in two different execution modes:

• Interrupt: the CSS (switch) handler signals a task that a message is incoming.
• No Interrupt: the CSS handler does not signal an incoming message, and hence may cause deadlock if a large number of messages is exchanged.

The current version of PVM on the SP-2 is built on top of the private implementation of MPI by IBM. We also must note that MPI is implemented not on top of MPL, but at the same level (on top of a common low-level library).