TCP Sequence Number

In 4.4BSD (and most Berkeley-derived implementations) when the system is initialized the initial send sequence number is initialized to 1. This practice violates the Host Requirements RFC. (A comment in the code acknowledges that this is wrong.) This variable is then incremented by 64,000 every half-second, and will cycle back to 0 about every 9.5 hours. (This corresponds to a counter that is incremented every 8 microseconds, not every 4 microseconds.) Additionally, each time a connection is established, this variable is incremented by 64,000.

Each endpoint of a TCP connection establishes a starting sequence number for packets it sends, and sends this number in the SYN packet that it sends as part of establishing a connection.

There is no requirement for either end to follow a particular procedure in choosing the starting sequence number. The operating system is free to use any mechanism it likes, but generally it's best if it chooses a random number, as this is more secure.

From that starting point, each packet sent by either end contains two sequence numbers - one to specify where in the stream the packet is, and an ACK sequence number which signifies the number of bytes received. Both numbers are offset by the starting sequence number.

Read all about it in Wikipedia of course - look for "sequence number" in that page to get all the gory details.