Networks require physical connections between computers or nodes, and there are several options when it comes to network topology. Early networks often employed a bus topology, in which nodes are connected to a single cable which acts as a backbone. A terminating resistor at either end of the cable prevented signal reflection. This topology had many advantages. It was simple to implement, and could function even if one or more nodes went down. There were several disadvantages, however. A break at any point in the cable, or a missing terminating resistor, could bring down the whole network.

Another problem stemmed from the protocol used on these early bus networks. Most of these networks used Ethernet, a technology that's commonly used for networking, and which defines the cabling used, the structure of network data (the frame), and the means of sending data. Early Ethernet networks used a method of sending data called carrier sense multiple access with collision detection (CSMA/CD). This method entails the following: a node that has a frame ready for transmission waits until the medium is idle. When the medium is idle, the node transmits a frame. The problem is that other nodes might also be transmitting frames at the same time. Therefore, each node that transmitted a frame must monitor for a collision. If a collision is detected, the transmitting node must resend the frame. The node will remain idle for a random period of time (thus avoiding a situation in which two or more nodes are constantly resending frames and causing collisions). If the medium is still idle after this random idle period, the node resends the frame. If not, it waits until the medium is idle to resend.

Such a method solved the problem of ensuring that data eventually got to its destination, but as you might have guessed, collisions on such networks are unavoidable, and the number of collisions increases as more nodes are added. As a result, competing technologies emerged to address this issue. The most popular of these was IBM's token ring network. A token ring network has a ring topology, although the ring was a logical one; the physical network topology was that of a star, with nodes connected to a device called a Multistation Access Unit (MAU) that functioned like a ring. A special frame, known as a token, traveled around the ring, and in order for a node to send a frame, it must seize the ring. This eliminated the problem of collisions. When IBM first introduced token ring hardware, the standard had a speed of 4 megabits per second (Mbps). Although this would, at first glance, place it as a disadvantage of Ethernet, which by then had been upgraded from 2.94 Mbps to 10 Mbps, since token ring networks had no collisions, they were able to achieve throughput that was comparable to or even better than Ethernet. The introduction of the 16 Mbps token ring technology in 1988 gave it even more of an advantage.

In spite of this, token ring struggled to achieve a significant market share, and in the end, Ethernet was able to prevail by providing more speed. Fast Ethernet, the designation given for Ethernet standards that provided speeds of 100 Mbps, was introduced as 100BASE-T. This standard required either Cat-3 or Cat-5 twisted-pair cabling, and instead of using a physical bus topology, all nodes connected to a hub and thus 100BASE-T had a physical star topology (the hub still functioned as a logical bus, however). The issue of collisions was solved by replacing hubs (which simply re-transmits frames to every port on the hub except the port on which the frame was received) with switches, which only transmit the frame to the destination device. Switches use a store-and-forward mechanism to send data, and each port on a switch forms its own collision domain.

Other layer 2 devices include repeaters and bridges. Repeaters are unintelligent devices that re-transmit network signals, allowing us to effectively increase the total range of a network. For example, if we have two 100Base2 networks, each of these networks can span a total distance of 185 meters. By putting a repeater between the two network segments, we could combine them into a single network and effectively double the size of the network to 370 meters.

Bridges, which can also be used to connect network segments, are more intelligent devices. Like a repeater, a bridge enables us to connect multiple network segments. However, a bridge has a forwarding database that it uses to determine whether a frame gets forwarded to another port on the bridge. At first, the database is empty, and when the bridge receives a frame, it floods all ports on the bridge with the frame. Over time, however, the bridge learns which nodes are connected to each port (or more accurately, connected to the segment that is connected to a port), and only bridge then only sends the frame to the correct port.