Half-duplex or full-duplex?
Full-duplex links are the key to extending the maximum network diameter of Fast (100 Mbps) Ethernet. Full-duplex requires separate receive and transmit paths (link segments consisting of no more than two devices). These devices can be Ethernet adapters or switching hub ports. Notice that we did not mention repeating hub ports. A repeating hub is part of the collision domain and reinforces collisions received on any of its other ports. A repeating hub is not capable of full-duplex operation. Although it is possible to have just two Ethernet adapters configured for full-duplex, expansion beyond two adapters requires a switching hub capable of supporting full-duplex operation.
Half-duplex means transmitting and receiving over the same medium but not at the same time. Full-duplex allows for simultaneous sending and receiving. Coaxial-based transceivers such as 10Base-5 and 10Base-2 are not able to invoke full-duplex since they do not have separate receive and transmit paths. However, 10Base-T and 10Base-FL do have separate switching hubs. If these interfaces are configured for half-duplex, then the simultaneous detection of receive and transmit activity will trigger collision detection. These same interfaces configured for full-duplex would disable this collision detection logic since full-duplex does not follow the CSMA/CD rules of shared Ethernet.
It is very important that a full-duplex link be configured properly. A station or switching hub port will send out frames at will, ignoring the CSMA/CD protocol of shared Ethernet, if it is configured for full-duplex. If the other end is configured for half-duplex, it will incorrectly detect collisions and take actions that could cause late collisions which are not automatically re-sent) and CRC errors. The result is a general slowdown of the network negating the benefits of migrating to Fast Ethernet.
As mentioned before, at 100 Mbps the maximum network diameter is short because of the limited collision domain at this speed. This is not a problem with twisted-pair link segments and switch ports because the maximum twisted-pair segment length is 100 m which is within the collision domain limit. The problem is with fibre-optic ports which allow segment lengths of 2 km for multimode operation and 15 km or greater for single-mode operation. Under the rules for half-duplex CSMA/CD Ethernet, our point-to-point fibre-optic segment is limited by the collision domain to 412 m. However, with full-duplex operation, which ignores the CSMA/CD algorithm, fibre-optic segments can be extended to their limit.
With Fast Ethernet, the use of switch technology is recommended. When using Fast Ethernet over fibre-optic cabling, full-duplex operation is recommended.
With the proliferation of Fast Ethernet and the similarity of the cabling components to conventional Ethernet, a means was proposed in IEEE 802.3u to automatically configure Fast Ethernet ports to work with either legacy Ethernet ports or other Fast Ethernet ports. This configuration protocol was based upon National Semiconductor's NWay standard. There is a way for twisted-pair links to automatically configure compatible formats in order for links to begin communicating. This scheme was intended for twisted-pair links and not coaxial buses. Coaxial cable is a legacy 10 Mbps standard that is not in the plans for evolving Ethernet. Fibre optics is a different story. Although fibre-optics is very much in the plans for evolving Ethernet, there is no simple way for two fibre-optic devices to auto-negotiate data rates since a 10Base-FL device operates at 850 nm while a 100Base-FX device operates at 1300 nm. These devices will not interoperate. However, there is nothing in the Auto Negotiation protocol to prevent two fibre-optic devices to auto-negotiate if communication is possible. Recognising this, the 100Base-SX standard was recently introduced which incorporates 850 nm fibre-optic components that can function at either 10 or 100 Mbps. At 100 Mbps, these devices are limited to 300 m segment lengths. Therefore, it is important that the installer fully understands the capabilities of the fibre-optic equipment. Frequently with fibre optics, the data rate is fixed and not negotiated. The auto-negotiation protocol functions best on twisted-pair links.
The benefit of auto-negotiation is to provide hands-free configuration of the two devices attached to the link segment. At connection time, each of the two devices will advertise all their technical abilities. These abilities have been ranked by the standard as shown in Table 1. The lowest possible ranking is 10Base-T which assumes half-duplex or shared Ethernet operation. The very next ranking is 10BaseT full-duplex indicating that full-duplex has higher performance than half-duplex. Finally, the highest ranking is 1000Base-T full-duplex. This ranking scheme has been provided for completeness. It is not assumed that a particular adapter can handle all technologies. In fact, some of these technologies may not have been commercialised. However, they are all listed consistent with the IEEE 802.3 standard.
Table 1. Auto-negotiation assumes a ranking of priorities. 10Base-T is at the bottom
Each device examines each other's technical abilities and determines the lowest common denominator. For example, if an Ethernet adapter can only handle 10Base-T while a switch port can handle either 10Base-T or 100Base-TX, 10Base-T will be chosen by both. If two Ethernet adapters connect, one only advertising 10Base-T and the other only advertising 100Base-TX, there will be no subsequent communication since no compatibility exists.
Auto-negotiation can be very helpful or it can be a source of problems especially in the area of half, full-duplex selection since it is difficult to ascertain what was selected. Usually a switching hub and adapters have indicators that will denote Fast Ethernet selection; however, there is usually no indication for half, full-duplex operation.
Part V will continue this discussion and look at the Transport Layer Protocols.
For more information: Jaap Willemse, Electronic Products Design, 012 665 9700, firstname.lastname@example.org