Packet switching
Also known as: IP packets, datagram, connectionless, circuit switching, TCP/IP
Packet switching is the network architecture that underlies the internet.
Instead of reserving a dedicated circuit between sender and receiver (as phone calls traditionally did), data is broken into packets that independently navigate the network, arriving at the destination to be reassembled. Multiple conversations share the same network infrastructure simultaneously.
Packet switching, explained at three levels
At home
When you send an email, the internet doesn't reserve a direct line from your computer to the recipient. Instead, the email is broken into small packets that each find their own way through the network, possibly taking different routes. They're reassembled at the destination. This is packet switching.
At work
Packet switching allows thousands of conversations to share the same network cable or fibre simultaneously. Each packet carries its destination address; routers forward it toward the destination. When a link is busy, packets wait in a queue. This is fundamentally different from phone calls, which reserved a dedicated circuit for the duration of the call.
At home
IP packets have a header (source address, destination address, TTL, protocol) and a payload (data). Routers forward packets based on the destination IP using routing tables: each router makes an independent decision. TCP (Transmission Control Protocol) handles reliability at a higher layer: it sequences packets, detects losses, and requests retransmission. UDP provides no reliability, leaving it to the application.
At work
Packet switching advantages over circuit switching: statistical multiplexing (when one user is idle, others use the capacity), resilience (packets route around failed links), and efficiency (no wasted reserved bandwidth). Disadvantages: variable delay (jitter), possible packet loss, and out-of-order delivery that upper layers (TCP) must handle. Real-time applications (VoIP, video) require additional QoS mechanisms to manage jitter and loss.
At home
The internet uses connectionless packet switching (each packet routed independently). Connection-oriented packet switching (like X.25, Frame Relay, ATM, and MPLS) establishes a virtual circuit first, combining benefits of both models. TCP's sliding window flow control and congestion control algorithms (CUBIC, BBR) manage sending rate to avoid overwhelming intermediate routers.
At work
Modern network QoS uses DSCP (Differentiated Services Code Point) markings in IP packet headers to classify traffic. Network equipment queues packets by class: voice packets in a priority queue with guaranteed bandwidth, bulk transfers in a best-effort queue. WRED (Weighted Random Early Detection) drops some lower-priority packets before queues fill, providing early congestion signals to TCP connections without dropping all packets at once.
Real-world examples
At home
Sophie downloads a large file while simultaneously streaming music. Both transfers share the same internet connection simultaneously: possible because packet switching allows the connection to carry packets from both streams interleaved, rather than dedicating the entire pipe to one at a time.
At work
Thomas's company's network team uses DSCP marking on their voice-over-IP system. Voice packets are marked EF (Expedited Forwarding) and placed in a priority queue, ensuring they're transmitted before bulk file transfer packets, maintaining call quality even during heavy network utilization.
How it compares
| feature | Packet switching | Circuit switching |
|---|---|---|
| Resource reservation | None (shared) | Dedicated per connection |
| Path | Per-packet (may vary) | Fixed for duration |
| Latency | Variable (queuing) | Fixed, predictable |
| Efficiency | High (statistical multiplexing) | Low (idle circuits waste capacity) |
| Used in | Internet, IP networks | Legacy phone networks, SONET |
For analysts
Understanding packet switching explains why network performance varies: the shared nature of packet networks means congestion affects all users simultaneously, which is directly relevant to capacity planning and SLA discussions.
History & origin
Packet switching was independently developed by Paul Baran at RAND Corporation (1964) and Donald Davies at the NPL in the UK (1965). Baran wanted a network with no single point of failure that could survive a nuclear attack; Davies wanted computers to share lines efficiently. ARPANET (1969), the precursor to the internet, implemented packet switching. The alternative, circuit switching (used in telephone networks), became obsolete for data networks. The internet's scalability and resilience are direct consequences of packet switching.