How can one analyze the efficiency of exchange schemes between branches of a parallel algorithm?

Question

Parallel Computing

There are 5 exchange schemes between the branches of the parallel algorithm: differentiated, translational, translational-cyclic, pipelined-parallel and collector. It is necessary to analyze their effectiveness, but I do not understand how at all. Are there any indicators of such efficiency? And it seems to me that the efficiency will vary, depending on the topology of the system and the implementation of the circuit itself.