Ethernet works by connecting an almost unlimited number of devices to the same cable. Usually, a host can just transmit whenever the cable is not is use. In the relatively uncommon case where two devices start transmitting at the same time, a collision occurs. Both senders then wait a random amount of time before transmitting again. In any case, every device on the cable can receive every packet, but discards all those not addressed to it. This scheme, one of many that can regulate access to a hardware medium, is referred to CSMA/CD, an acronym for Carrier Sense, Multiple Access / Collision Detect.
There has been much debate over Ethernet performance. The performance of any CSMA/CD network will depend on several considerations, including the method of determining silent times after a collision, the length of the cabling, the size of packets, and the amount of traffic. The Ethernet standard defines how silent times are determined, and the network engineer can seldom influence this feature anyhow. The remaining factors are summarized in the table below, though be aware that the standard places constraints on several of these.
|Performance Factor||Desirable Feature|
|Cable length||Short||Short cables reduce the chance of collisions, since electrical signals take less time to propagate between hosts|
|Packet size||Large||Large packets reduce the chances of a collision, since collisions can only occur during a fixed time window at the beginning of a packet|
|Amount of traffic||Light |
(< 20% capacity)
|More traffic means more collisions; for standard 10 Mbit/s Ethernet, try not to exceed 2 Mbit/s on any single segment|
In the early 1980s, IEEE began developing standards for LAN technologies, including Ethernet. Unfortunately, the new standard, IEEE 802.3, is not compatible with the older DIX specification. Due to Ethernet's already large installed base, both the newer IEEE standard and the persistent DIX pseudo-standard are in use.
Ethernet operates at 10Mbps per second, and its packets are between 64 and roughly 1500 bytes in length. A 6-byte address is used, which is divided into a 3-byte vendor ID and a 3-byte vendor-defined field. Ethernet manufacturers are assigned a unique vendor ID, and are then responsible for insuring that all of their devices have unique addresses in the last 3 bytes. An all-ones address is interpreted to mean broadcast, and all hosts attached to the cable process such a packet. A group of addresses are assigned for multicast use, and most newer Ethernet cards permit software to select a subset of the multicast addresses for reception.
The IEEE Ethernet Standard is not available online, because the sale of standards documents are a major income source for IEEE and other standards organizations.
Ethernet encapsulation standards describe how IP packets should be converted into Ethernet packets. ARP (Address Resolution Protocol) describes how IP addresses are translated to Ethernet addresses.