Broadly speaking, there are two types of switched networks in use: circuit switched and packet switched. A circuit switched network is defined by a physical or virtual circuit (or connection) that connects two endpoints and has a certain circuit bandwidth. A circuit only needs to be defined for the duration of the message transfer on a circuit switched network. Because switching devices can be used to redefine different connections, a circuit switched network can be reconfigured as needed.

The penultimate circuit switched network is the Public Switched Telephone Network, or PSTN. When you place a phone call to another party, a circuit is created between the two of you for the duration of the call. Circuit switched networks are data networks as well as voice networks. Another example of a circuit switched network is ISDN (Integrated Services Digital Networks). The best way to think of what a circuit switched network does is to remember that circuit switched networks are stateful. Stateful means that you can define a message transfer in terms of:
1. A source
2. A destination
3. A path of the circuit
4. A cost for the path based on time, performance, or some other weighting

In a circuit switched network, you can represent nodes as a graph in graph theory, connections as weighted edges between nodes, and the actual defined or preferred paths through the graph, which are called routes. In real terms, messages are sent from endpoint to endpoint as a complete unit. If you have multiple IP packets (or datagrams), they all travel down the same route on a circuit switched network.

A packet switched network is based on a different concept, that of the best available route. On a packet switched network, individual packets are sent from a source to a destination by the best connection available at the switching device. This type of network is designed for inherently unreliable networks where connections are transient. If a connection drops out, the next packet is sent to a different next hop. A packet switched network cannot guarantee a path. A certain percentage of packets will reach a dead end where, as they say in Vermont, “You can’t get there from here,” and so some packets will get dropped or returned. Packets will also arrive out of sequence.

Therefore, packet switched networks require a mechanism to ensure that all lost packets are resent
and that packets can be sequenced to retrieve the data that they encode. Of course, the prototypical packet switched network is the Internet, or more broadly speaking, networks based on the Internet Protocol. Other networks that are packet switched are X.25, Frame Relay, Asynchronous Transfer Mode (ATM), and Multiprotocol Label Switching (MPLS), among others.

The best way to think of what a packet switched network does is to remember that packet switched networks are stateless. Stateless means that you can define a message transfer in terms of:
1. A source
2. A destination
3. The position of the packet in the sequence
4. A Time-to-Live (TTL) for the packet, which may be based on a hop count or timeout parameter, and is the time after which the packet expires and is dropped at the next device that receives it.

Circuit switched networks have their advantages and disadvantages over packet switched networks. A circuit switched network sends an entire message over the same circuit, which can be faster than sending parts of a message over many paths. When a message arrives, the data arrives in sequence and doesn’t need to be reassembled. By contrast, a packet switched network makes better use of the network’s capacity because it can distribute traffic over many connections. The extra overhead involved to sequence incoming packets and the loss of performance is offset by the more efficient use of the network and the much higher fault tolerance offered by packet switching. Neither model, whether circuit or packet switched, is better than the other; they are simply different.

What both circuit switching and packet switching have in common is that they both have switches that can change the network’s topology. To understand modern networks, you need to understand how switches operate. Switches not only control the physical connections between network segments through electrical connections, but different classes of switches also have the intelligence to measure the performance of different paths, determine routes, and optimize the preferred paths or routes to nodes on the network in a stored but dynamic table. Internetworks and WANs would not
function without the use of these types of routing devices.