The question at this point is whether that conclusion applies to three
or more processes. That is, can the general two-process analysis be
applied on a message-by-message basis to all send/receive messages
between any two processes in a multi-process computation and, more
importantly, give the desired overhead compensation result? We look
at two scenarios with three processes to get a
sense of the answer. These scenarios are:
| Pipeline | Process P1 sends a message to P2, then P2 sends to P3 | |
| Two Receive | Process P1 and P3 sends a message each to process P2 | |
We argue that these two scenarios are enough to elucidate all similar cases regardless of the number of processes. Again, we follow a rational reconstruction approach to determine approximated executions and then derive expressions for updating overhead, waiting time, and delay variables to match the reconstructed executions. Only the Two Receive scenarios is described in detail in this paper.