What Is Packet Switching?

Packet switching is a method of moving data across a network in which each message is broken into small blocks called packets. Each packet carries a header with its source and destination address and a sequence number, followed by a slice of the data, the payload. Network devices forward every packet on its own, and the receiver reassembles them into the original message.

The idea was developed independently in the 1960s by Paul Baran at the RAND Corporation and Donald Davies at the UK National Physical Laboratory, who coined the word packet. ARPANET, the forerunner of the internet, put it into service in 1969. The Internet Protocol, defined for IPv4 in RFC 791, is a packet switching protocol, and so is Ethernet, which calls its packets frames.

A packet switched network lets many conversations share the same links, because packets from different senders are interleaved on the wire. Capacity is used only when someone has data to send, which is far more efficient for bursty computer traffic than reserving a line for each conversation.

Every packet goes through the same steps, whether it crosses one switch or a continent.

Circuit switching sets up a dedicated path between the two ends before any data is sent and holds it for the whole call, as the traditional telephone network did. Packet switching reserves nothing in advance.

Packet networks have largely replaced circuit networks, even for voice, because sharing links is so much cheaper. The trade-off is variable delay, which matters for voice, video and control traffic.

Packet switching comes in two forms. In datagram switching, used by IP, each packet is routed independently and packets from one message can take different paths and arrive out of order. In virtual circuit switching, used by older X.25 and Frame Relay networks and by MPLS today, a path is set up first and every packet of the flow follows it, which keeps packets in order and makes traffic easier to engineer.

An Ethernet switch forwards frames using a MAC address table, either storing each frame and checking it before sending (store and forward) or starting to send as soon as it has read the destination (cut-through). Our explainer on how an Ethernet switch forwards traffic covers both, and our comparison of layer 3 switches and routers covers forwarding by IP address.

Four delays add up at each hop: processing, while the device reads the header; queuing, while the packet waits for the link; transmission, the time to put the bits on the wire; and propagation, the time the signal takes to travel. Queuing delay is the one that varies, and its variation, called jitter, is the main weakness of packet switching for real-time traffic.

Control networks deal with it in two ways. Quality of service marks time-critical packets so switches send them first. Time-Sensitive Networking goes further, scheduling traffic so delivery times are bounded, and our guide to Time-Sensitive Networking explains how. On the plant floor, managed industrial switches provide the priority queues that keep control traffic ahead of bulk data.

Industrial networks once used dedicated serial links and fieldbuses, which behaved more like circuits: one master, fixed polling and predictable timing. Moving that traffic onto Ethernet brings it onto shared, packet switched links alongside camera streams, file transfers and remote access.

That works well when the network is designed for it. Separate control and bulk traffic with VLANs, give control packets a higher priority, keep switch uplinks well below full load, and avoid chaining so many switches that the delay at each hop adds up. Where a process needs guaranteed timing rather than priority, TSN switches schedule the traffic instead.

We supply managed industrial switches with the priority queues and VLANs a control network needs, and configure them with the computers and devices in every system we build.