Bridges, Switches and Switching Hubs

Basically the terms bridge, switch and switching hub designate the same. In the early beginning of the ETHERNET, the term bridge was formed by the fact that a bridge had only two network connections. Later, so-called multiport bridges with several connections came up which were also called switches or switching hubs. This is why we use the common term “switch” in the following descriptions for all the components mentioned above.

The use of a switch is another variant of connecting network segments to each other. The decisive difference between a hub and a switch is that a switch is operating on the second layer of the ISO/OSI model, the MAC layer.

The following sections describe the functionality of such a layer 2 switch. For reasons of completeness it has to be mentioned that switches operating on higher and therefore protocol-specific layers also exist.

Using a switch, load separation between networks can be implemented which leads to an increased performance due to the reduced load of the individual segments. In contrast to a hub, a switch does not operate transparently (i.e. it doesn’t forward all data packages via all ports) but decides on the basis of the MAC target address whether and via which port an incoming data package has to be forwarded. The data package is only forwarded if the target station is located in another segment or if the target address of the data package contains a multicast or broadcast address.

As already mentioned, the decisive advantage of a switch is the logical separation of networks. Therefore, a switch represents a border for a collision domain. Aside from the performance improvement, the use of a switch allows a network to be extended beyond the usual borders.

Use of hubs and switches

To enable crosswise traffic between the segments, a switch has to be able to temporarily store the incoming data packages until they can be transmitted on the forwarding segment. The decision about forwarding of data packages is done using address tables. These address tables are generated by the switch itself during a self-learning process. During this process, the switch remembers the source addresses (MAC addresses) of incoming data packages of a port. If it later receives further data packages, the switch compares their target addresses with the entries in the address tables of the ports and, in case of a match, forwards the respective package via the corresponding port. Here, the following cases have to be distinguished:

• If the source station and the target station are located within the same segment, the data package is not forwarded.
• If the station of the target address is located in another segment than the source station, the data package is forwarded to the target segment.
• Data packages containing a multicast or a broadcast address as the target address are forwarded via all ports.
• A data package with a target address which is not contained in the address tables is forwarded via all ports (Frame Flooding).

The latter case normally only occurs during the first time after starting a switch since the address is usually entered after some time when exchanging a data package.

In order to limit the size of the address tables, addresses which are not used over a longer period of time are additionally removed from the tables. This also avoids incorrect forwarding as it would appear e.g. when a station is moved within the network.

To enable the building of a redundant network structure (as it is often found in more complex networks) using switches, the so-called spanning tree method has been introduced. With this method, the switches exchange configuration messages among themselves. This way the optimum route for forwarding data packages is determined and the creation of endless loops is avoided. The exchange of messages is performed cyclic. As a result a connection breakdown is detected and forwarding is automatically changed to another route.

Redundant network structure using only switches

Using a switch instead of a hub increases the bandwidth of the individual segments and therefore leads to an increased performance. Building a network consistently with switches furthermore enables full duplex operation and thus simultaneous data traffic in both directions since switches are able to establish dedicated peer-to-peer connections between the individual ports. The use of the access method CSMA/CD is not required since collisions can no longer occur. Depending on the network structure, this can further increase the performance drastically. For full duplex connections furthermore no length restrictions of the collision domain have to be observed.