When connecting Ethernet segments together to build your network, hubs (or repeater hubs) are typically used. In a star topology, a hub is necessary to connect several point-to-point segments in a hub-and-spoke fashion. However, attaching several devices to a hub simply “shares” the network. All devices attached transmit and receive over the same “shared” network, and therefore contend for the same medium and bandwidth. As more devices communicate on the network, performance may diminish as more collisions occur. In a repeater hub network, all attached devices are said to be in a single collision domain.
A network switch, on the other hand, provides for independent collision domains on each port. This means that within the switch, processor technology is used to accept an incoming Ethernet packet on one of its ports, determine the destination port, and forward the message to that port only. This is in contrast to a hub that simply rebroadcasts any incoming message out on all ports.
As an example, in a small network with two computers, six Ethernet-based brains, and an eight-port switch, each device and its corresponding switch port represent a single collision domain. Therefore, each device has access to the full bandwidth on the segment (no sharing of bandwidth), without contending with other devices (no collisions). This results in dramatically higher performance and near-deterministic response times. Furthermore, most switches allow for full-duplex operation, resulting in an effective doubling of throughput (that is, 10 Mbps becomes 20 Mbps, and 100 Mbps becomes 200 Mbps) because transmits and receives occur at the same time over separate twistedpair wires. When a SNAP Ethernet device is connected to such a switch, full-duplex, 200 Mbps overall throughput is fully supported and attainable.