13.2.3 Packet Switching

Packet switching combines the efficient resource sharing of message switching with the responsiveness required for interactive communications. Instead of transmitting an entire message as one large unit, the message is divided into many smaller packets, each of which is transmitted independently through the network. Because packets from many different users can be interleaved on the same communication links, packet switching makes highly efficient use of available bandwidth and has become the dominant switching technique for modern data networks, including the Internet.

Each packet consists of two parts: a header, which contains control information such as addressing and routing data, and a payload, which carries part of the user's information. At the receiving end, the packets are reassembled to reconstruct the original message.

By dividing messages into relatively small packets, individual links remain occupied for only short periods before packets from other users are transmitted. This statistical sharing of network resources allows many users to communicate simultaneously while making efficient use of transmission capacity. It also reduces transmission delay compared with message switching because intermediate switching nodes need only receive an individual packet before forwarding it to the next node.

Three principal approaches to packet switching are commonly used.

Datagram networks. In a datagram network, every packet is treated as an independent unit and is routed individually through the network. Each packet contains sufficient addressing information for network switches (routers) to determine its destination and select the most appropriate route. Because different packets belonging to the same message may follow different paths, they may arrive at the destination out of order or, occasionally, not arrive at all. The receiving device is therefore responsible for reordering packets, detecting missing packets, and requesting retransmission if necessary. The Internet Protocol (IP), which forms the foundation of the Internet, is the best-known modern example of a datagram service. Datagram networks are highly resilient because packets can be rerouted automatically around failed or congested links without requiring the entire communication session to be re-established.
Virtual-circuits. A virtual circuit (VC) establishes a logical path through the network before data transmission begins. Unlike a physical circuit, the communication links are not reserved exclusively for one user; instead, packets from many virtual circuits are statistically multiplexed over the same physical links. Once established, all packets belonging to the virtual circuit normally follow the same route and therefore arrive in sequence. This simplifies packet reassembly and can provide more predictable performance. Virtual circuits may be either permanent or switched, the latter being established only for the duration of a communication session. Technologies such as X.25, Frame Relay, and Asynchronous Transfer Mode (ATM) employed virtual circuits extensively, while modern Multiprotocol Label Switching (MPLS) networks use similar connection-oriented concepts within the core of many service-provider networks to support traffic engineering, quality of service (QoS), and virtual private networks (VPNs).
Hybrid networks. Many modern communication networks combine the advantages of both approaches. To end users, the network appears to provide a simple connectionless datagram service using the Internet Protocol. Internally, however, service providers may employ connection-oriented mechanisms such as MPLS label-switched paths to optimize routing, improve reliability, guarantee quality of service, and manage network traffic efficiently. This combination provides the flexibility of datagram networking while retaining many of the operational advantages of virtual circuits within the network core.

Packet switching offers numerous advantages over both circuit switching and message switching. Network resources are shared efficiently, communication links are utilized only when packets are being transmitted, and traffic can be rerouted automatically around failures or congestion. These characteristics make packet switching particularly well suited to bursty traffic such as web browsing, email, cloud computing, video streaming, file transfer, and Internet of Things (IoT) communications. The principal disadvantages are increased protocol overhead, variable transmission delay, delay variation (jitter), and the possibility of packet loss during periods of heavy network congestion. Modern communication protocols employ buffering, flow control, congestion control, error detection, and retransmission techniques to minimize the effects of these impairments.

The development of packet switching represented one of the most significant advances in the history of communications. By combining efficient statistical multiplexing with flexible routing and robust error recovery, it provided the foundation for the global Internet and virtually every modern data communication network. One specialized form of packet switching, known as cell switching, further refines these concepts by transmitting information in fixed-length packets, or cells, to provide more predictable switching performance for high-speed networks.