Note: Many topics at this site are reduced versions of the text in "The Encyclopedia of Networking and Telecommunications." Search results will not be as extensive as a search of the book's CD-ROM.

Network switches are multipoint connection devices that provide a point of attachment for a single computer or another device (hub or switch) that has multiple computers attached to it. The most important feature is that any device attached to one port can directly communicate with a device on another port over what is essentially a private link.

The significance of this technology can be seen when compared to older Ethernet shared LAN technologies. The traditional coaxial cable Ethernet LAN implements a linear cable topology that is shared by all the computers attached to it. Only one device can transmit at a time, so some computers will need to wait while another is transmitting. Computers "listen" for signals on the cable to see if it is being used.

Network switching reduces or removes the sharing of the network and the problems that result from sharing, such as contention (when computers wait to use the cable), collisions (when two systems attempt to use the cable at the same time), and delays caused by contention and collisions. See "Collisions and Collision Domains" and "CSMA/CD (Carrier Sense Multiple Access/Collision Detection)."

Figure S-13 illustrates the evolution of switching. In (A), all the workstations are attached to a single shared segment and must deal with the problems described previously. Many of these problems are solved via segmentation, as shown in (B), where the network is divided and rejoined with a bridge. This create two "collision domains" but retains a single broadcast domain. A collision domain is a group of network nodes that are contending for the same shared communication medium. Contention occurs in collision domains when two or more stations attempt to use the medium at the same time. By splitting the network, contention and collisions are reduced because there are fewer nodes per segment.

In (C), the linear network is replaced with a hub to create a star topology. This is basically the same network as (B), but the network topology is a star configuration. In (D), the switch provides multiport bridging. Each port provides dedicated bandwidth to the device attached to it. Any port can be bridged to any other port so that an end-to-end private link can be established between any two devices. Note that when two end stations are connected together, they contend with one another for access to the channel. A full-duplex link solves this by providing a dedicated wire pair for sending and receiving.

Notice that hubs can be attached to a single port, which means that all the stations attached to the hub share the same switch port. The assumption is that the users attached to the hub don't generate much network traffic, except possibly between other users on the same hub.

This model in (D) may be used on a small department level or as the topology for entire enterprise networks. It is extremely scalable. Switched networks do not lose performance as new switches are added, assuming higher-level switches provide adequate bandwidth. See "Network Design and Construction" for information about building hierarchical networks with switches.

You'll hear a lot about multilayer switches, layer 3 switches, layer 4 switches, and so on. These are "enhanced switches" with added functionality such as routing (thus, it's a multiport router) and packet inspection functionality that can provide prioritization and QoS. A switch with routing functionality is sometimes called a "layer 3 switch" or a "router switch." They are described under the topic "Multilayer Switching."

Since basic switches are layer 2 devices, networks built with them are flat, meaning that they are not subnetted into separate routing domains. Multilayer switches support routing. However, networks that are constructed with basic switches can be virtually separated into separate routing domains by using VLAN technology. See "VLAN (Virtual LAN)."

Many switches are now optimized for load balancing. They inspect and distribute traffic across multiple channels and to multiple devices (servers and storage). Packets are forwarded based on layer 4 and layer 7 information, including information at the application layer, such as the requested URL, cookies, or even processing requirements. For example, a database request is forwarded to a database server and Web requests to Web servers. Forwarding may also be based on persistent sessions in a case in which a server is already servicing a session and packets related to the same session should go to the same server. See "Load Balancing."

This topic continues in "Network Design and Construction," which describes how to design and build enterprise networks with switches.

Copyright (c) 2001 Tom Sheldon and Big Sur Multimedia.
All rights reserved under Pan American and International copyright conventions.