The first Ethernet was developed at Xerox in 1976 by Metcalfe and Boggs, based on a 3Mbps serial cable bus, called the Ether, and used 1-persistent CSMA/CD for medium access. In 1980 a consortium consisting of Xerox, DEC and Intel produced the Ethernet Specification, an updated version whose primary goal was to stabilise the standard at a higher speed of 10Mbps. The Ethernet Specification, defined what was called Ethernet II, and was used as the basis for IEEE 802.3, which is actually a generalisation covering a family of similar systems. There are minor differences between the MAC protocol used in Ethernet II and IEEE 802.3 but IEEE have provided a mechanism to allow for backward compatibility in the form of the SNAPextension to LLC. By the early 1990s Ethernet/802.3 had become the dominant LAN technology but it was becoming clear that increases in bandwidth requirements made a faster version desirable. IEEE began to investigate possibilities of a 100Mbps Ethernet and in fact produced two alternatives: 802.3u, which is now universally known as Fast Ethernet, and 802.12, so-called 100VG-AnyLAN, which, although theoretically superior, is less compatible with the existing 802.3, and has proved much less popular. 802.3u has experienced widespread success largely because of its ease of interoperability with older 10Mbps equipment and is now the norm for modern workstation PCs. By the late 1990s technology allowed another tenfold increase in speed to gigabit Ethernet (1000Mbps) standardised as IEEE 802.3z/802.3ab . A 10,000Mbps Ethernet has been standardised under the auspices of IEEE working group 802.3ae.
Strictly, the term “Ethernet” refers to a specific product, which predated the general standard 802.3. However, conventional usage has blurred this distinction, and any 802.3 compliant network is now commonly referred to as an “Ethernet”, as indeed is any network which uses a MAC sublayer similar to that specified in 802.3, even when other physical media or enhancements to the MAC protocol have been introduced. In the original Ethernet the broadcast medium was coaxial cable, forming a single-channel baseband serial bus operating at 3Mbps or 10Mbps. 802.3 now also allows the use of twisted-pair and optical fibre in addition to "Ethernets" based on switches rather than conventional broadcast media.
Coaxial cable is still sometimes used for Ethernet implementations, but has now largely been replaced with twisted pair technology. Two main types of coaxial cable have been employed: 10mm 10Base5, known as thick Ethernet, allowing up to 100 nodes over a maximum length of 500m (in a single segment); and 5mm 10Base2, or thin Ethernet, supporting only 30 nodes over 185m. In the twisted pair Ethernet, 10Base-T, nowadays far more prevalent, a coaxial cable Ether is no longer used at all, but replaced with a backplane-based box called a hubto which each station is attached via two twisted pair data cables, one for each direction (Figure 1). The hub collapses the functionality of the Ether into a single small unit, and all long-distance signal propagation is over the point-to-point twisted pair links. The hub repeats any signal received on an input to all its other outputs, emulating the broadcast feature of the cable bus. A station can detect a collision very simply by watching for simultaneous activity on its input and output pairs.
Hubs have several advantages over extended cable buses. Firstly, they allow for centralised monitoring, maintenance and upgrading of the network infrastructure. Secondly, they can use unshielded twisted pair (UTP) wiring already present in many modern office buildings. Thirdly, a break or fault in a cable does not bring the whole network neighbourhood down as in cable based Ethernet. However, UTP runs must be no more than about 100m in length, which is somewhat more restrictive than thick Ethernet. Physically a hub configuration is a star rather than a bus topology, but from the MAC point of view this makes no real difference, as the hub is also using CSMA/CD. Like the cable it replaces, the total bandwidth available in a standard Ethernet hub is 10Mbps and this must be shared between all attached stations.
Figure 1: Typical 10Base-T configuration
The performance of Ethernet hubs can be enhanced if their design is modified so that rather than relying on a single backplane, multiple paths can be introduced through the device, allowing more than one transfer to occur simultaneously, so long as they are to different destinations. This breaks away from the basic CSMA/CD principle, and effectively turns the LAN into a switched network. However, this Ethernet switch is interfaced to the stations by the same UTP connections as 10Base-T, and therefore, to these stations, looks just like a 10Base-T hub, but with superior performance. Obviously the total bandwidth provided by an Ethernet switch must be much greater than that of a hub. Collisions can still, occur, if, for example, the switch tries to send data to a station while it is beginning a transmission, but this assumes that the connection between switch and station is half-duplex. However, with a switch, it also becomes possible to use the two twisted pairs connecting to the station as full duplex 10Mbps connections, although this does require some augmentation of the original Ethernet or 802.3 specifications, as these have no support for full-duplex operation. Building on work by a number of manufacturers on an unofficial specification, full duplex switched Ethernet or FDSE, IEEE has now generated a full-duplex standard, 802.3x which covers full-duplex extensions to 802.3 at any speed.
In full duplex operation, the concept of a "collision" as an event detected by a station, no longer exists. If several stations try to transmit to the same target at the same time the switch will transmit only one of the contending packets and will buffer the rest. Of course ultimately, the switch may run out of buffer space, whereupon it must throw away new packets, relying on upper layers for recovery. 802.3x includes an optional MAC Control sublayer which defines new control frames which are carried inside conventional 802.3 frames: the only control function supported at present is flow control which allows a switch to use a Pause frame to request that a station inhibit further data transmission for a specified period of time. This flow control mechanism helps reduce the necessity for packet discard but its implementation is optional. This Ethernet switching bears little resemblance to the original CSMA/CD concept and in this form Ethernet is a switched network, relying on standard switching principles. Note that a switch can also be employed to connect segments of an Ethernet together, each based on its own hub. The switch then routes between these segments.
All the functions required at a station including support for full duplex operation can nowadays easily be combined onto a small number of integrated circuits usually mounted on an Ethernet network interface card (NIC), which can plug into a standard bus on a PC or printer.