Consider the following three causes of a 1-second network delay between A and B. In all cases, assume ACKs travel instantly from B back to A.

(i) An intermediate router with a 1-second-per-packet bandwidth delay; all other bandwidth delays negligible

(ii) An intermediate link with a 1-second propagation delay; all bandwidth delays negligible

(iii) An intermediate router with a 100-ms-per-packet bandwidth delay, and a steadily replenished queue of 10 packets, from another source (as in the diagram in 14.2.4 Example 4: cross traffic and RTT variation).

How might a sender distinguish between these three cases? Hint: consider packet pairs (14.2.6 Packet Pairs).

Following just an information of above problem

14.2.6 Packet Pairs

One approach for a sender to attempt to measure the physical bandwidth of the bottleneck link is the packet-pairstechnique: the sender repeatedly sends a pair of packets P1 and P2 to the receiver, one right after the other. The receiver records the time difference between the arrivals.

Sooner or later, we would expect that P1 and P2 would arrive consecutively at the bottleneck router R, and be put into the queue next to each other. They would then be sent one right after the other on the bottleneck link; if T is the time difference in arrival at the far end of the link, the physical bandwidth is size(P1)/T. At least some of the time, the packets will remain spaced by time T for the rest of their journey.

The theory is that the receiver can measure the different arrival-time differences for the different packet pairs, and look for the minimum time difference. Often, this will be the time difference introduced by the bandwidth delay on the bottleneck link, as in the previous paragraph, and so the ultimate receiver will be able to infer that the bottleneck physical bandwidth is size(P1)/T.

Two things can mar this analysis. First, packets may be reordered; P2 might arrive before P1. Second, P1 and P2 can arrive together at the bottleneck router and be sent consecutively, but then, later in the network, the two packets can arrive at a second router R2 with a (transient) queue large enough that P2 arrives while P1 is in R2s queue. If P1 and P2 are consecutive in R2s queue, then the ultimate arrival-time difference is likely to reflect R2s (higher) outbound bandwidth rather than Rs.

Additional analysis of the problems with the packet-pair technique can be found in [VP97], along with a proposal for an improved technique known as packet bunch mode.