When a host receives an IP packet, the host's networking stack looks at the Protocol field to determine to which upper-layer protocol to send the data. If the Protocol field in the IP header is 6, the data contains a TCP segment. If it's 17, then it contains a UDP datagram. Consequently, a single host can use the same UDP and TCP port numbers simultaneously.
Transmission Control Protocol
RFC 793 somewhat redundantly describes TCP as a “connection-oriented, end-to-end reliable protocol.” TCP provides the following features:
Reliable delivery
Ordering
Error control
Flow control
Congestion avoidance
The phrase “connection oriented” refers to TCP's attempt to simulate the properties of a physical connection. Sound familiar? An ideal connection provides reliability and order. The data the sender sends is exactly what the receiver receives and in the same order.
TCP guarantees both order and reliability through the use of sequence numbers and acknowledgments. The sender marks each TCP segment with a sequence number that increments in a predictable fashion. This allows the receiver to reassemble the segments in the correct order should they arrive out of order. The receiver sends an acknowledgment (ACK) in response to each segment, indicating the sequence number of the segment it received. If the sender fails to receive an ACK after a period of time, it will retransmit the segment.
Connection Establishment
You should be familiar with TCP's famous three-way handshake. Its purpose is to allow both sides to establish initial sequence numbers. Here's how it works:
1 Node A sends Node B a segment with the synchronize (SYN) flag set, along with its initial sequence number (ISN), which is usually random.
2 Node B responds with a SYN and its own ISN, which is different from Node A's ISN. Node B also sends an ACK to acknowledge Node A's initial SYN.
3 Node A acknowledges Node B's SYN by sending an ACK.
Connection Termination
When the client is ready to terminate the connection, it sends a segment with the finish (FIN) flag set. The server responds with an ACK; however, it can continue to send data. This is called a half-close state. When the server is ready to terminate the connection, it too sends a segment with the FIN flag set. The client ACKs this segment, and the connection is fully closed.
Connection Reset
The TCP Reset (RST) flag is useful for cleaning up old and duplicate connections. Generally, when a TCP receives a segment for a port that it's not listening on, it will respond with a RST, letting the sender know that the connection was refused. This can happen if the sender tries to send data over an old connection that it believes is still open.
In another case, if a TCP doesn't receive any ACKs from a host, it will abort the connection by sending a RST. This may happen if there's a unidirectional link or someone abruptly shuts down the host, precluding it from gracefully terminating the connection. On the off chance that the host receives the RST, it will abandon the connection and may attempt to open a new one.
Error Control
Error control detects whether application data was lost or corrupted during transit. Aside from port numbers, this is the only feature both TCP and UDP have in common. Both TCP and UDP provide error control by calculating a checksum over the entire packet and including it in the header. The recipient calculates its own checksum over the entire packet and compares it to what's in the header. If they don't match, the recipient discards the packet.
Flow Control
Flow control prevents a sender from sending data faster than the receiver can receive it. TCP provides flow control through the use of windows and acknowledgments. The receiver specifies a window—the number of bytes that the sender may send before stopping to wait for an acknowledgment from the receiver.
Congestion Avoidance
Congestion avoidance can be described as TCP's politeness algorithm. If a sender fails to receive an ACK or receives duplicate ACKs, TCP assumes that there's congestion somewhere in the network and slows down the rate at which it sends.
Encapsulation and Decapsulation
The entire process of moving data from one application to another over the network rests on two processes: encapsulation and decapsulation. Without encapsulation and decapsulation, there is no network. In short, each time a higher-layer protocol passes data down to a lower-layer protocol, it's encapsulated. Whenever a lower-layer protocol passes data up to a higher-layer protocol, it's decapsulated.
Encapsulation and Multiplexing
Although they're technically different, in networking parlance multiplexing and encapsulation are used interchangeably. Multiplexing is the act of sending multiple streams of data over a single connection. Nodes achieve multiplexing by using encapsulation, as shown in Figure 1.7.
Figure 1.7 Encapsulation of a TCP segment and IP packet inside an Ethernet frame
As an example, suppose a TCP-based client application needs to send some data to a server. The application instructs the host operating system to open a TCP connection to the server. Once the connection is open, the application uses it to send some data to the server. Here's what happens: The host's operating system creates a TCP segment, places the application data in the Data field, and populates the rest of the fields in the TCP header appropriately. The OS encapsulates the TCP packet inside an IP packet, with the Protocol field set to 6 (TCP). It then sends the IP packet to the Ethernet driver, which encapsulates the IP packet in an Ethernet frame, which has the Type field set to 0x0800 (IP). The Ethernet driver passes the frame to the network interface, which converts the data to bits and puts it out on the wire.
Decapsulation and Splitting
Splitting or demultiplexing is the multiplexing process played backward. Continuing with the preceding example, when the server receives the Ethernet frame containing the IP packet from the client, the Ethernet driver looks at the Type field in the Ethernet header and sees that it's 0x0800, indicating an IP packet. The Ethernet driver decapsulates the IP packet from the frame and passes it to the operating system's IP stack. The IP stack looks at the Protocol field and sees that it's 6, indicating that the Data field contains a TCP segment.
Summary
The backstory of the OSI model and Ethernet seems to have little relevance to “modern” networking. But upon closer inspection, many of the fundamentals of networking haven't changed in 50 years. Essentially, the goal of networking is and always has been to enable applications to transfer data. The challenge of networking is to make that happen without having