How TCP Delivers Reliably

TCP promises a complete, ordered stream over a network that loses and reorders packets, and a handful of mechanisms keep that promise. Seeing them at work explains the delays every TCP application lives with: the round trip before the first byte, the slow start of a new connection, and the stall when one segment goes missing.

It begins with sequence and acknowledgement numbers, which everything else is built on, and follows them through the handshake. Two limits on the sender come next, flow control and congestion control, and a comparison sets them side by side. An example follows one lost segment until it is sent again, a walkthrough shows how a connection closes, and the last ideas cover head-of-line blocking and why reusing a connection saves time.

The bit-level layout of every header field and the many congestion algorithms in use are left aside.

By the end you can order the handshake, compute an acknowledgement number, tell the two kinds of control apart, predict head-of-line blocking, and say why a connection pool makes requests faster.