Note: Many topics at this site are reduced versions of the text in "The Encyclopedia of Networking and Telecommunications." Search results will not be as extensive as a search of the book's CD-ROM.

A packet is a unit of data that is transmitted across a packet-switched network. A packet-switched network is an interconnected set of networks that are joined by routers or switching routers. The most common packet-switching technology is TCP/IP, and the Internet is the largest packet-switched network. Other packet-switched network technologies include X.25 and IPX/SPX (the original Novell NetWare protocols). This topic focuses on the Internet and TCP/IP packet-switched networks.

The concept of a packet-switched network is that any host connecting to the network can, in theory, send packets to any other hosts. The network is said to provide any-to-any service. The network typically consists of multiple paths to a destination that provide redundancy. Packets contain header information that includes a destination address. Routers in the network read this address and forward packets along the most appropriate path to that destination.

A simple packet-switched network is shown in Figure P-1. If computer A needs to send a packet to computer Z, the packet first travels to R1. By looking in a routing table, R1 determines that the best path to the destination is through the R2 interface. The network has a redundant topology; so, if the link between R1 and R2 fails, a path through R3 and R4 will reach R2.

Routers run routing protocols to discover neighboring routers and the networks attached to them. These protocols let routers exchange information about the network topology as it changes due to new or failed links. See "Routers" and "Routing."

The Internet and all IP networks are fundamentally connectionless datagram networks. A datagram is a packet formed at the network layer in the Internet protocol suite. In this context, a packet is either a full datagram or a fragment of a datagram (a datagram that has been split into multiple smaller pieces to comply with size restrictions on some networks). A connectionless network provides no end-to-end delivery guarantees. No connection is set up to track and guarantee packet deliveries, nor do network routers maintain any sort of state about a packet flow between sender and receiver (virtual circuits).

However, TCP does provide these services if they are needed. End systems can use TCP to set up connections and track packet deliveries. The recipient uses TCP to acknowledge packet receipt. Unacknowledged packets are retransmitted by the sender. See "Reliable Data Delivery Services" and "TCP (Transmission Control Protocol)" for more information.

David P. Reed, one of the people involved in the early design of the TCP/IP protocols, discusses the importance of having both a connectionless protocol layer (IP) and a connection- oriented layer (TCP) in his paper "The End of the End-to-End Argument." By splitting the Internet protocols into two protocols (IP and TCP), the network was "decentralized" and turned into a basic datagram forwarding network. In this model, end systems implemented functions that were previously handled by networks, such as flow control, acknowledgments, and retransmissions. The decentralized model also allowed end users to deploy applications in their computers. Compare this to the telephone network, which relies on the telephone company to provide applications. Reed's paper is at (http://www.reed.com/Papers/endofendtoend.html).

A comparison between the telephone system and the Web is interesting. The user interface for the telephone system is a 12-key pad, while the user interface for the Web is a fully customizable and extensible Web browser interface. The telephone company provides limited applications (call-waiting, called id, and so on), while the Web allows users to deploy their own applications in their end systems.

The advantage of the connectionless packet model is that packets are forwarded independent of other packets. Packets are forwarded on-the-fly by routers, based on the most current best path to a destination. If a link or router fails, packets are quickly diverted along another path. Since routers don't maintain information about virtual circuits, their job is greatly simplified. In contrast, ATM and frame relay networks are connection oriented. A virtual circuit must be established between sender and receiver across the network before packets (cells or frames, respectively) can start to flow. One reason the Internet has scaled so well is that there is no need to build virtual circuits for the millions of flows that cross the network at any one time. Routers simply forward packets along the best path to the destination. However, in the interest of speed and QoS, virtual circuits are being implemented on the Internet by using protocols such as MPLS (Multiprotocol Label Switching).

Packets have a header and a data area. The header holds address and routing information. Think of a packet as an envelope in which the destination address is written on the outside of the envelope and data goes inside. The process of building packets is covered under "Network Architecture."

A single transmission may require hundreds or thousands of packets-for example, a large file is broken up into many small pieces that are inserted in the payload area of packets. This scheme helps overcome transmission problems. If a glitch occurs, only one packet may be affected. Then it is only necessary to retransmit that one packet rather than the entire file.

In relation to the OSI protocol model, packets are formed in the network layer and passed down to the data link layer, where they are encapsulated into the frames of the underlying network. Frames cross a single point-to-point link between network devices, while packets cross multiple router-connected links. In other words, frames are isolated to a single link, while packets are envelopes for delivering data across internetworks. Packets are broken up into frames for delivery across a network; but, when the frames reach the next router, the packet information is examined by the router and a decision is made about how to forward the packet across the next link.

Packet-switched networks use multiplexing principles. Packets from multiple sources can traverse links and routers in an interleaved fashion. In fact, a single host can establish multiple simultaneous sessions that transmit packets across the same link. For example, you can open two Web browsers and connect to two different Web sites at the same time. The packets from both connections are interleaved across the link. When compared to dedicated leased-line circuits, packet-switched networks use bandwidth efficiently. The network is shared by many users who generally keep the pipe full. Leased lines, in contrast, use TDM (time division multiplexing), which can waste bandwidth by reserving time slots for data even when there is no data to send.

The Lucent/Bell Labs Web site has a good animation of how packet multiplexing works. Go to http://www.bell-labs.com/technology/network/packetidea.html.

Packet-switched networks have been called networks of queues. Packets arriving on a link are pushed into buffers and queued up for processing. The router figures out where to forward the packets and pushes them into an appropriate outgoing queue. The problem is finding the right buffer size. Small buffers drop packets, while large buffers may cause excessive delays as packets wait in line for processing (input) or transmission (output). A single host can saturate a router with a burst transmission, blocking other users and causing congestion. A number of mechanisms to control congestion and improve quality of service are discussed under "Congestion Control Mechanisms." Also see "Flow-Control Mechanisms," "Prioritization of Network Traffic," "QoS (Quality of Service)," "Queuing," and "Traffic Management, Shaping, and Engineering."

Historically, packet networks have been an irritation to the telecommunications companies, who made big money by leasing circuits. ATM was an attempt to push cell relay as the predominant internetworking technology. However, datacom engineers advocated variable-size packets and packet switching, rather than ATM's fixed-size cells and virtual circuit requirements. The outcome of this rift was summarized by Charles N. Judice writing in IEEE Communications Magazine, August 2000:

Recently, Internet engineers have been concerned with the loss of transparency on the Internet. This is the Internet concept of a single universal logical addressing scheme and the mechanisms by which packets may flow from source to destination essentially unaltered. RFC 2775 (Internet Transparency, February 2000) explains that end-to-end transparency has been lost due to the deployment of NATs (network address translators), as well as firewalls, proxies, and caches. These devices cause problems for Internet applications that require stable and continuous IP addresses. In such cases, custom application-level gateways are required to perform translation for those applications; but even then, end-to-end transparency may not be restored.

Recently, a new type of packet called a "SmartPacket" has been defined for Active Networks. SmartPackets carry both data and user-specified methods that can be used to control switches in the network. SmartPackets support rapid customization of the network. The programs carried by packets are executed at each node visited by the packet. These programs may include diagnostics, monitoring, and automatic configuration utilities. See "Active Networks."

The following Internet RFCs provide additional information about packet networks:

• RFC 791 (Internet Protocol, September 1981)


• RFC 793 (Transmission Control Protocol, September 1981)


• RFC 970 (On Packet Switches with Infinite Storage, December 1985)


• RFC 1180 (A TCP/IP Tutorial, January 1991)

Copyright (c) 2001 Tom Sheldon and Big Sur Multimedia.
All rights reserved under Pan American and International copyright conventions.