Transmission Control Protocol (TCP)

TCP is the connection-oriented transport protocol in the TCP/IP protocol suite. TCP attempts to make connections reliable through the use of positive acknowledgement with retransmission – a system where acknowledgements are required for all data sent. If the sending host does not receive these acknowledgements, retransmission eventually occurs. Sequence numbers are assigned to segments to be sure that they are assembled in the correct order at the receiving system. Flow control is handled through buffering techniques such as the use of sliding windows.

When two systems wish to communicate via TCP, they first establish a session (remember the 3-way handshake?), also known as a connection or virtual circuit. Once data has been transmitted and the session is complete, it is closed via a modified 3-way handshake – the difference is that the connection needs to be closed (and acknowledged) independently by both systems. The figure below outlines the 4 steps required to close a TCP session. Closing a session involves sending FIN (no more data to send) and ACK messages. Notice that the extra step involved – the system that originated the end of the connection acknowledges that the other system is now closing the connection as well.

Figure: Ending a TCP session.

There can be a delay between the time when one system attempts to close a connection and the time that the entire connection is actually closed. This occurs because one end may close the connection, while the application on the other end might still be running. A good example is when a user accesses a website – the user may keep their browser running long after the web server has finished sending the requested data as has attempted to close the connection.

TCP communicates with upper-layer applications and protocols through the use of port numbers. When combined with an IP address, a port number defines a unique endpoint for communication. There are 65,536 port numbers available to TCP. How they are chosen requires a bit of discussion.

For the most part, server-side applications are defined by what are considered to be well-known port numbers. These port numbers are defined in RFC 1700 and fall into the 0-1023 range. For example, when a client opens a web browser and types in http://www.2000trainers.com, that name will be translated into an IP address using DNS, and then a connection will be made to port 80 on that IP address. TCP port 80 is the default port on which a web server waits for a connection. But what about the client side of the connection? Well, the client needs to specify a port on which it will send and receive data as well. These port numbers are usually in the range above 1024, and are dynamically assigned by the source host.

Author: Dan DiNicolo

Dan DiNicolo is a freelance author, consultant, trainer, and the managing editor of 2000Trainers.com. He is the author of the CCNA Study Guide found on this site, as well as many books including the PC Magazine titles Windows XP Security Solutions and Windows Vista Security Solutions. Click here to contact Dan.