20.9 Summary

As we said early in the chapter, there is no single way to exchange bulk data using TCP. It is a dynamic process that depends on many factors, some of which we can control (e.g., send and receive buffer sizes) and some of which we have no control over (e.g., network congestion, implementation features). In this chapter we've examined many TCP transfers, explaining all the characteristics and algorithms that we could see.

Fundamental to the efficient transfer of bulk data is TCP's sliding window protocol. We then looked at what it takes for TCP to get the fastest transfer possible by keeping the pipe between the sender and receiver full. We measured the capacity of this pipe as the bandwidth-delay product, and saw the relationship between this and the window size . We return to this concept in Section 24.8 when we look at TCP performance.

We also looked at TCP's PUSH flag, since we'll always see it in trace output, but we have no control over its setting. The final topic was TCP's urgent data, which is often mistakenly called "out-of- band data." TCP's urgent mode is just a notification from the sender to the receiver that urgent data has been sent, along with the sequence number of the final byte of urgent data. The programming interface for the application to use with urgent data is often less than optimal, which leads to much confusion.