A TCP/IP network can have a lot of traffic flowing across it at any given time. In large networks the flow of information can be too much, causing bottlenecks and congestion that essentially bogs the network down to the point it is ineffective. To alleviate this, a network will be divided into smaller networks called subnets.

Broadcast traffic
Broadcasting is the nature of TCP/IP traffic. When a router receives a packet with a destination address it does not know, the packet is broadcast everywhere on the network.

Simply put, a broadcast will occur when a router receives a packet, looks at the destination, realizes it does not know where to find this destination, and asks the other routers, "Do you know this address?" One router will respond, and a route for that address will be created.

The portion of the network that can receive the broadcast is called a broadcast domain. In a large network the broadcast domain is the entire network. If each computer on the network is sending packets, this can cause electronic gridlock fairly quickly because a copy of each packet is sent to every computer on the network even if it was not meant to receive it.

Hubs and switches
Networks used to be connected by hubs. A hub simply repeats all the signals it receives from one port to all of its other ports, which is known as broadcasting. To solve the broadcast problem engineers made the hubs smarter so they could learn which computers were on the network and send packets to only the computers they were meant for, logically dividing the network into smaller pieces to limit broadcast traffic. The smart hubs were called switches.

This approach worked well for small and medium sized networks but in larger networks there was still a problem. While a switch is learning it acts like a hub, and because it only remembers the computers on a network for a limited time, a large network with infrequent traffic would require a switch to constantly relearn. In this scenario, the switch is no better than a hub.

Removing the bottleneck
As an extension of the switch, which logically divides a network, a large network can be manually divided into smaller networks so that broadcasting is limited to the small sections. These smaller networks are called subnets.

Subnets are created by configuring the IP addresses for all the computers in the subnet to be similar to each other, but different from other subnets. The different subnets are kept separate by using a subnet mask. A subnet mask filters IP addresses allowing computers with specific IP addresses to talk to each other directly yet other computers will not hear their broadcast traffic. With a large network divided into many smaller networks, how can they all be connected so they could talk to each other?

Backbone and routers
A backbone is a high capacity, high bandwidth connection that is used to connect the many subnets of a large network. If we connect all the subnets together how will they communicate without broadcasting traffic again? The backbone uses routers to limit broadcast traffic to the subnet while still connecting computers in different subnets together. A router is an intelligent device that has the ability to find a path from a computer in one subnet to a target computer in a different subnet and connect the two without any broadcast traffic from the subnet leaking out into the backbone.

By using subnets, backbones, and routers it is possible for a large network to operate efficiently without any bottlenecks or congestion.