Ethernet Hardware
The Ethernet standard includes the specification of the kind of cables needed. Originally, a thick coaxial cable was required, but over the years there have been many updates.
This is a selection of Ethernet cable types:
| Name | Cable | Max Segment | Speed | |
| 10Base5 | Thick coax | 500 m | 10Mb/s | 'Thicknet', chunky yellow, vampire taps |
| 10Base2 | Thick coax | 200 m | 10Mb/s | 'Cheapernet', 'thin ethernet', 'thinnet', actually 185 m max segment |
| 10Base-T | Twisted pair | 100 m | 10Mb/s | Telephone wire, Cat 3 cable; 150 m on Cat 5 |
| 10Base-F | Fibre optic | 2000 m | 10Mb/s | Long distance |
| 100Base-T4 | Twisted pair | 100 m | 100Mb/s | 100Mb Ethernet on two telephone cables (Cat 3) |
| 100Base-T | Twisted pair | 100 m | 100Mb/s | 100Mb Ethernet, Cat 5 cable, aka 'fast Ethernet' |
| 100Base-F | Fibre optic | 2000 m | 100Mb/s | Long distance |
| 1000Base-T | Twisted pair | 100 m | 1Gb/s | Gigabit Ethernet, Cat 5 or Enhanced Cat 5 |
There are several others, including optical fibre gigabit. The Max segment is the longest allowable stretch of cable in a single unbroken section.
The most important aspect is the increase of speeds in the standards. The 100Mb and 1000Mb (aka Gigabit) Ethernet standards are more or less the same as the original standard (in terms of frames layout, CSMA/CD, etc.) but simply faster. There are one or two wrinkles, such as minimum frame size and the amount of time you listen for a collision, but mostly it just follows through.
In terms of deployment, 10Mb is currently in decline, 100Mb is very common, 1Gb is widespread, 10Gb is emerging and talks on 40Gb are progressing.
The original Ethernet, 10Base5, runs on a fat coaxial cable with vampire taps containing a transceiver. The taps are little boxes that bite into the Ether cable to make a connection. There is a minimum gap of 2.5 m between taps: Figure below. The transceiver contains the electronics to do the collision detection. This supports a cable (up to 50 m long) to the host containing five twisted pairs of wires, with a large and unwieldy socket (AUI socket) on the end.
There are many rules governing 10Base5 that are necessary to ensure good function. A segment can be extended with up to a maximum of four repeaters (which simply amplify the electrical signal), therefore five segments (total length of 2460 m) can be connected together. Of the five segments only three can have devices attached (100 per segment). A total of 300 devices can be attached on a single thicknet broadcast domain. This is called the 5-4-3 rule.
However, 10Base5 is expensive, inflexible (physically as well as metaphorically), hard to install, and the AUI plugs have a tendency to fall out of their sockets; 10Base2 and 10Base-T are much simpler.
Thinnet, or 10Base2, uses simple coaxial cable, much like TV aerial cable. It plugs in using simple passive BNC (variously called British Naval Connector, Bayonet Neill Concelman or Bayonet Nut Connector) connectors (no vampire taps or transceivers needed). This is much cheaper and easier to lay around the building and to connect up. The necessary electronics are now on the interface card in the machine, not in a separate transceiver. The minimum gap between connections is 0.5 m and a segment can have up to 30 machines. The BNC plugs directly into the card in the computer with no intervening cable.
A segment can be extended with other segments using up to four repeaters, i.e., five segments in total. Two of these segments, however, cannot have hosts and can only be used for extending the length of the network (to 925 m). What this means is that three segments with a maximum of 30 stations on each can give you 90 devices on a thinnet broadcast domain.
Twisted pair connects in a very different way. Also known as unshielded twisted pair (UTP), this is literally a bunch of eight wires twisted together in four pairs: the twists help reduce electrical interference. This is essentially a bundle of high-quality telephone cables. On the ends are RJ45 (Registered Mike) plugs.
UTP has a standard classification into levels of quality according to the ability to transmit signals:
- Category 1: No performance criteria.
- Category 2: Rated to 1 MHz (used for telephone wiring).
- Category 3: Rated to 16 MHz (used for Ethernet 10Base-T).
- Category 4: Rated to 20 MHz (used for Token Ring, 10Base-T).
- Category 5: Rated to 100 MHz (used for 1000Base-T, 100Base-T, 10Base-T).
- Enhanced Category 5: Rated to 200 MHz (used for 1000Base-T, 100Base-T, 10Base-T).
- Category 6: Rated to 250 MHz (used for 1000Base-T).
Now 10Mb Ethernet requires the use of Cat 3 (or better) cable, whereas 100Mb needs Cat 5. You can run Gigabit Ethernet on Enhanced Category 5. This is good-quality Category 5 where you need to be really careful about making good, clean connections in the plugs and sockets and is rated to 200 MHz. Category 6 supports Gigabit with ease. The next category, Category 7 (600 MHz), is in the process of standardization.
Further, in an Ethernet, instead of a single backbone cable that all the machines connect to, each machine is wired to a central box. This can be a hub or a switch, both of which relay packets from one segment to another. The difference between them is the degree of intelligence in the forwarding.
A hub is ignorant of any protocol above the physical layer and simply echoes all packets onto all wires. This makes a hub-connected system look electrically like a 10Base2 or 5 Ethernet. It is a single collision domain as a packet from any host has the potential to collide with packets from all other hosts, regardless of their source and destination. The collision domain shares the 10Mb or 100Mb available bandwidth.
A switch understands a link layer protocol, e.g., Ethernet. It knows where each host is and is able to direct a packet down that single segment that has the right machine on the end rather than copying it to all the output wires. This means now that each output cable is a separate collision domain. There will only be a collision if two machines try to send to the same destination at the same instant; there will be no collision between machines sending packets to different machines at the same instant. Thus there is potentially 10/100Mb bandwidth available between each pair of machines simultaneously.
Often switches are even more cunning than this: they can store and forward packets. If a switch finds the output channel is busy it can store a packet in internal memory and forward it later when the channel is free. This means that each source-destination path is a separate collision domain. If this is happening we can do away with CSMA/CD completely, as there can be no collisions! If the output channel is exceptionally busy, the memory buffer can fill up and then the switch will just drop any more incoming packets until space becomes available. Recall that Ethernet CSMA/CD will itself drop packets after 16 tries, so this is not so bad as it seems. It is up to a higher layer protocol to discover the loss and take remedial action.
Incidentally, it is this point that is the major cause of lost packets in the modern Internet: being dropped through congestion en route causing buffers to overflow. An alternative to store and forward is cut through switching. In this, the switch starts forwarding the frame before all of it has been read. This decreases the latency of the packets as they travel through but reduces the overall reliability as the switch might be forwarding a corrupt packet: it cannot know if a packet is corrupted until it has been completely read and the checksum computed.
Switches are more complicated (read 'expensive') than hubs, but give better total throughput and better efficiency under high load (few collisions). These days even switches are cheap and it is getting increasingly difficult (and pointless) to find a hub. Hubs and switches (Figure below) are also applicable to the other kinds of Ethernet, but are overwhelmingly associated with UTP.
If you have a switch and your hardware supports it, twisted pair can run full duplex, meaning that there can be 10/100Mb to the machine simultaneously with 10/100Mb from the machine. This is a total of 20/200Mb per pair of machines in flight at a time. This is because there can be no collisions between inward and outward traffic as they are transported over different twisted pairs within the cable. The alternative is half duplex, where traffic can only go in one direction at a time. Typically, half-duplex Ethernet uses CSMA/CD while full-duplex Ethernet is switched and does not.
You can buy converters that connect together the different kinds of Ethernet cable, e.g., thinnet to twisted pair. Some network cards have more than one type of socket on them. Thicknet is virtually never seen these days, with thinnet only found in old installations and UTP being widely used, and Cat 5 (or better) UTP being used for all new installations.