In the echo test, node A sends to node B, node B sends the message back to A. A measures the roundtrip time for N repetitions and for different size messages.
In the exchange test, node A sends to B at the same time node B sends to A, both nodes receive the respective messages and repeat the sequence N times.
In the sink test, node A just sends N back to back messages to B. B just receives them as fast as it can. Both A and B time the sequence. B starts its timer when it receives the first message. The transmitter often overruns the receiver. The transmitter's time often cannot be trusted. (The OS may be buffering the transmissions.) Unfortunately, if packets are lost, the receiver may not complete and report its times, unless there is out-of-band signaling to mark the end of the transmission.
Message buffers should be paged in. Buffer alignment can effect performance as can message length, not just the total length, but an odd number of bytes may be handled less efficiently than a multiple of two or four etc.