A hub floods every frame to everyone. The fix is a device that learns where each host is and sends a frame only there. That single idea — learn, then forward — is what turns a hub into a switch.
The repeaters-and-hubs article ended on the hub's flaw: it scales the wiring, but it floods every frame to every host, which brings no privacy, constant collisions, and bandwidth shared across everyone.
The goal now is to keep the convenience of one central device while stopping traffic from going where it does not belong. That requires something a hub fundamentally lacks — knowledge of where each host is. The first device to have it was the bridge.
A bridge sits between two sets of hub-connected hosts. It has only two ports, one facing each side.
Its new ability is to learn which hosts are on which side, simply by watching the traffic that passes through it. (The exact learning mechanism — reading the source address of every frame — is the switch's core trick, taken apart in a later article.)
Knowing which side each host is on lets the bridge contain traffic:
This is a genuine turning point. A repeater and a hub are address-blind — they move signals with no idea who those signals are for. A bridge is the first device to read a frame's Layer 2 (MAC) address and make a decision based on it. That is the jump from moving signals to moving frames based on their destination.
It also pays off in a way the collision-domain idea from the previous article predicted: by refusing to forward same-side traffic, the bridge splits one collision domain into two. Each side now only collides within itself, so collisions drop. That reduction is the bridge's real reward.
A switch is the bridge idea generalized. Instead of two ports separating two sides, it has many ports — and it learns which host is on each individual port.
So a switch combines both earlier devices: like a hub, many hosts plug into it; like a bridge, it learns where hosts are and contains traffic. The difference is granularity — per port, not per side.
When two hosts communicate, the switch knows exactly which two ports they sit on and passes the frame only between those two. Every other port sees nothing at all.
That per-port precision unlocks three things a hub could never offer:
Private, collision-free, full-speed links to every host — that is why the switch completely displaced the hub.
These four devices are not four unrelated inventions. They are one idea, refined step by step:
device layer ports what it does with a frame
------ ----- ----- -------------------------
repeater 1 2 regenerate, send on
hub 1 many regenerate, flood to all
bridge 2 2 learn side, forward only if needed
switch 2 many learn per port, forward to one
Each row adds intelligence: first more ports, then the ability to read addresses, then per-port precision. The switch is where that line arrives.
Here is the switch's role stated exactly: a switch facilitates communication within a single network. Every host connected to one switch belongs to the same network — they share one IP address space, as the networks article described, and they can reach one another directly.
A handful of hosts on one switch, all addressed within 192.168.1.x, is precisely a typical home Wi-Fi network: same switch, same network, direct communication between every device on it.
But that is the switch's hard boundary. It works only within one network. A switch makes every decision using MAC addresses and never so much as looks at an IP address or a network. So when a host in one network needs to reach a host in a different network, a switch is helpless — it has no concept of "another network" to send toward. Crossing that boundary needs a different device entirely. That device is the router, and it is the next article.
In the real world, "bridge" is a word from the past — nobody buys a standalone bridge today. The reason is simply that a switch is a multi-port bridge: it implements the exact same learn-and-forward logic (standardised as 802.1D bridging), just with many ports instead of two. So when documentation calls a switch a "bridging device," or a hypervisor exposes a "network bridge" to connect virtual machines, that is not a different technology — it is this same idea wearing its original name. Bridge and switch are one concept at two scales.