Full-duplex Ethernet

While the early coaxial cable based variants of Ethernet were half-duplex by design, all the common variants of twisted pair (10BASE-T, 100BASE-TX and 1000BASE-T) and fiber optic Ethernet provide separate channels for send and receive.

To allow use of hubs and for compatibility with existing variants of Ethernet they were initially implemented in a half-duplex manner with the transceiver (usually by this point integrated into the device) detecting a collision if an attempt was made to transmit and receive simultaneously and looping back data to the host so it could hear itself transmit (as it would on a shared medium).

However if both ends of the link are not hubs, and the hardware supports it, the two channels can be split and used to make a full-duplex link. Unfortunately if autonegotiation is enabled on one end and forced full-duplex on the other, the end with autonegotiation will detect the link as half-duplex causing large numbers of errors due to duplex mismatch.