duplex
termWhether a link can carry traffic in both directions at once, or must take turns.
Modern switched links are full duplex and the setting is rarely thought about - until a mismatch, where one end negotiates full and the other assumes half. The result is the most misleading fault in networking: the link is up, small pings succeed, and throughput collapses under load with rising errors, because one side transmits while the other believes a collision occurred. It survives as a diagnostic reflex precisely because it looks like anything except what it is.
Ethernet began half duplex by necessity: on a shared medium only one station could transmit, and collision detection settled who went next. Full duplex became standard in 1997 with IEEE 802.3x, which added a 'full duplex mode of operation on a speed-independent basis' with pause-based flow control. It needs a link with only two ends, which switching provided, and on it collision detection is simply switched off.
From 10 Gb/s on, the question disappeared: every task force since 802.3ae has written the same objective, 'Support full-duplex operation only'. Duplex is now a property of 10, 100 and 1000 Mb/s copper links and of the auto-negotiation that sets them, which is where it still goes wrong.
Also known as: half duplex, full duplex