Packet switching is a method of data transmission in which a message is divided into small, self-contained units called packets, each carrying addressing information, and every packet is forwarded independently across shared network links from node to node until it reaches its destination. Among internet fundamentals it is the core principle on which the Internet Protocol and the entire internet are built.
Unlike circuit switching, which reserves a dedicated path for the whole duration of a call, packet switching uses statistical multiplexing: link capacity is shared dynamically among all traffic, and a link carries data only when someone actually has data to send. This makes networks far more efficient for bursty computer traffic, at the cost of variable delay and possible packet loss under congestion.
How packet switching moves data across a network
The sender breaks data into packets, each with a header containing source and destination addresses and control fields, and a payload carrying a slice of the data. Routers along the path read the destination address, look it up in a forwarding table and send the packet to the next hop. Each router decides independently, which is why packets of the same message can take different routes.
Most packet switches use store-and-forward operation: a router receives a whole packet, checks it, places it in an output queue and transmits it when the outgoing link is free. When packets arrive faster than a link can send them, the queue grows; when the buffer is full, further packets are dropped. Higher-layer protocols such as TCP detect loss and retransmit, and reorder packets that arrive out of sequence.
| Delay component | Cause | Depends on |
|---|---|---|
| Processing delay | Header inspection and lookup | Router hardware, usually microseconds |
| Queuing delay | Waiting behind other packets | Load on the link, from zero to many milliseconds |
| Transmission delay | Pushing bits onto the link | Packet size divided by link rate |
| Propagation delay | Signal travel time | Distance, roughly 5 microseconds per kilometer in fiber |
Packet switching compared with circuit switching
The classic telephone network used circuit switching, in which a call reserved a fixed channel through every exchange for its entire duration. That guaranteed constant quality but wasted capacity during silence. Packet switching trades that guarantee for efficiency and resilience.
| Property | Circuit switching | Packet switching |
|---|---|---|
| Resource reservation | Dedicated path for the whole session | None; links are shared |
| Setup phase | Required before data flows | Not required for datagrams |
| Use of capacity | Idle when the caller is silent | Used only when data is sent |
| Delay | Constant after setup | Variable, depends on queues |
| Failure handling | Call drops if a link fails | Traffic can be rerouted |
| Typical example | Traditional telephone network | Internet Protocol networks |
The efficiency gain comes from the fact that data traffic is bursty. A user browsing the web transmits in short bursts separated by long pauses. With statistical multiplexing, a link can serve many more users than its capacity would allow if each had a reserved channel, as long as their peaks rarely coincide.
Datagram and virtual circuit approaches
Packet-switched networks come in two main variants. In the datagram model, used by IP, each packet is routed independently and no state is kept per connection inside the network. In the virtual circuit model, a path is set up first and packets carry a short label identifying it, so switches forward them along the pre-established route.
| Variant | Path selection | State in the network | Examples |
|---|---|---|---|
| Datagram | Per packet, hop by hop | Routing tables only | IPv4, IPv6 |
| Virtual circuit | Once, during setup | Per-connection entries | X.25, Frame Relay, ATM |
| Label switching | Per flow, set by control protocols | Label tables in the core | MPLS in provider networks |
The internet’s choice of datagrams reflects the end-to-end principle: keep the network core simple and stateless, and place reliability, ordering and congestion control in the end hosts. This design let the internet scale across independent networks without central coordination of every flow.
Origins of packet switching
Packet switching was developed independently in the early 1960s by Paul Baran in the United States, who studied survivable communication networks, and by Donald Davies in the United Kingdom, who coined the word packet. The concept was first deployed at scale in ARPANET in 1969, followed by the NPL network and the French CYCLADES project, which influenced the datagram design later adopted by TCP/IP.
Commercial packet networks based on the X.25 standard spread in the 1970s and 1980s, but the IP datagram model eventually prevailed. Today even voice and video, the original strongholds of circuit switching, are carried over packet-switched IP networks.
Limitations and engineering responses
Sharing links means no built-in guarantee of bandwidth or delay. Real-time applications such as video calls suffer from jitter, the variation in packet delay, and from loss during congestion. Several mechanisms address these weaknesses without abandoning packet switching.
- Congestion control in TCP and QUIC slows senders when loss or rising delay signals full queues.
- Quality of service marking with DSCP lets routers prioritize latency-sensitive traffic.
- Jitter buffers in voice and video applications smooth out variable arrival times.
- Active queue management such as CoDel keeps router queues short and limits bufferbloat.
- Traffic engineering with MPLS steers flows onto paths with spare capacity.
A common misconception is that packets always take different paths. In practice routing is stable for long periods, and most packets of a flow follow the same route; per-packet path diversity is possible, not typical.
A plain-language summary
Imagine moving a large bookcase by mailing it in many small boxes. Each box gets its own address label. The boxes travel through the postal system on whatever trucks are free, and they are reassembled at the end.
Packet switching sends computer data the same way. Messages are cut into small labeled pieces that share the same roads with everyone else’s pieces. No road is reserved for one person. That is why the internet can carry traffic from billions of devices at once.
An example from practice
A company with offices in two cities replaces leased voice lines with internet-based telephony. Under the old setup each call reserved a 64 kbit/s channel for its full duration, even when nobody spoke, and the company paid for 30 such channels between the sites.
The new system sends voice as small packets every 20 milliseconds over the same IP link used for email and file transfers. To protect call quality, the network team marks voice packets with a high-priority DSCP value and configures routers to serve that queue first. Phones use a jitter buffer of about 40 milliseconds to absorb delay variation.
Call quality remains clear during busy office hours, and the company drops the separate voice lines entirely, because packet switching lets voice and data share one connection efficiently.