a network bridge
A bridge is the historical middle step between a dumb hub and a modern switch. Picture two crowded rooms joined by a single doorway with a clever doorkeeper: the doorkeeper only lets a message through the doorway if the person it is addressed to is actually in the other room. Conversation that stays within one room never crosses, so each room gets quieter. That doorkeeper is a bridge.
Technically, a bridge connects two or more LAN segments and operates at the data link layer. It reads the source MAC address of each frame to learn which segment a machine lives on, building a small table. When a frame arrives, it looks up the destination MAC: if the destination is on the same segment the frame came from, the bridge drops it (no need to forward); if it is on the other side, the bridge forwards it; if it has not learned the destination yet, it floods the frame to the other segments. This is exactly the MAC learning and forwarding logic later built into switches.
By keeping local traffic local, a bridge splits one collision domain into several, so machines on different segments can transmit at the same time without colliding. A switch is essentially a multi-port bridge implemented in hardware, with one segment per port. You rarely buy a standalone bridge today, but the concept is the direct ancestor of every switch — and the term lives on in software, for example a Linux bridge joining virtual machines.
Two old hubs, each with several PCs, are joined by a bridge. PC-A and PC-B on the left hub talk all day; the bridge sees both source MACs are on the left and never forwards their frames to the right hub, so the right-side PCs stay free to talk among themselves.
The bridge isolates local traffic, creating two collision domains from one.
A bridge breaks up collision domains but, like a switch, forwards broadcasts everywhere — so it does not break up the broadcast domain. Only a router or VLAN does that.