Half duplex, full duplex, and duplex configuration

Duplex communication means a channel carrying signals both directions (vs. simplex, one direction only). Ethernet has two duplex settings.

Half duplex

One direction at a time, never send and receive simultaneously. A walkie-talkie. Every device waits its turn to transmit, which hurts performance.

This is legacy territory: hubs. Collisions happen, so half duplex uses CSMA/CD (Carrier Sense Multiple Access with Collision Detection). When a collision is detected, the offending devices stop, each waits a random time, then retransmits. The random backoff makes a repeat collision unlikely.

Full duplex

Bidirectional: send and receive at the same time, like a phone call. Transmit and receive ride separate wire pairs, so frames from the two end nodes cannot collide, and the collision-detection circuit is disabled in full-duplex mode.

Requirements: point-to-point only, a dedicated switched port, and full-duplex support on both ends. All modern NICs (Ethernet, Fast Ethernet, Gigabit) support it.

Configuring duplex and speed

SwitchX(config)# interface fa0/1
SwitchX(config-if)# duplex full
SwitchX(config-if)# speed 100

The options are duplex full | half | auto and speed 10 | 100 | 1000 | auto. Auto is the default for both on Catalyst switches.

Common practice: auto toward PCs (their NICs negotiate fine), and hardcode switch-to-switch links on both sides so a failed negotiation or a rogue 10 Mbps device can’t downgrade the uplink.

Port-type rules

• 100BASE-FX: full duplex only, 100 Mbps only, cannot autonegotiate.
• 10/100/1000 ports: half or full at 10/100 Mbps, but at 1000 Mbps it’s full duplex only.

Duplex mismatch

The classic failure. When autonegotiation fails (the far side doesn’t support it), a Catalyst switch defaults the port to half duplex. If the far device is hardcoded half, fine. Hardcoded full: mismatch.

The symptom is late collision errors on the connection. The fix, and the prevention, is manually setting the switch to match the attached device. Mismatched settings on directly connected devices is the number one cause of duplex problems.

Verifying

show interfaces <name> (privileged EXEC) shows it all:

FastEthernet0/5 is up, line protocol is up (connected)
  ...
  Full-duplex, 100Mb/s, media type is 10/100BaseTX
  ...
  0 output errors, 0 collisions, 1 interface resets
  0 babbles, 0 late collision, 0 deferred

Duplex and speed on one line, and the error counters below. Nonzero late collision means go check for a mismatch.