Previous | Table of Contents | Next |
If one end of the link is a 10BASE-T system that does not support Auto-Negotiation, the partner is automatically configured for 10BASE-T half-duplex operation (default mode). When confronted with another networking technology that uses the RJ-45 connector (e.g., Token Ring), Auto-negotiation will automatically fail the link.
Auto-Negotiation is based on the link pulse used in 10BASE-T. For Auto-Negotiation, the link pulse is divided into 33 fast link pulses that are used to carry pages of coded information between link partners.
Full-duplex operation supports simultaneous signaling in both directions over dedicated links by turning off the CSMA/CD collision detection circuitry. It provides some increase in bandwidth over links that have a high proportion of bidirectional traffic, such as switch-switch and switch-server links. In addition, full-duplex operation increases the maximum length of fiber links. Whereas a half-duplex link is limited to 412 meters by the need to detect collisions, full-duplex operation supports links of up to 2 kilometers because no collision detection is required. This increased link length is only useful for fiber links, signal attenuation limits, and copper link length to 100 meters for both half- and full-duplex operation.
Flow control provides a method for controlling traffic flows between intermediate devices (primarily switches and routers) and between intermediate devices and servers to avoid dropping packets. Currently two-speed (10/100 or 100/1000) operation requires large buffers to reduce the probability of dropping packets when a continuous stream of packets is sent from a high-speed to a low-speed device (e.g., 100M bps to 10M bps, or 1000M bps to 100M bps). In such a scenario, when the buffers fill, the intermediate device drops the unbuffered packets.
Exibit 10. Repeater Connection Styles.
Flow control provides a management alternative to having large buffers. When a buffer approaches full, the receiving device can send a flow control packet back to the sending device to stop the incoming packet stream. When the buffers of the receiving device empty, packet transmission starts again. This eliminates dropped packets and allows manufacturers to build switches with smaller buffers, which reduces costs.
Repeaters provide for shared media operation in 10BASE-T and 100BASE-T via the CSMA/CD protocol. 10BASE-T networks have a collision domain diameter of 1000 meters. This permits building large, single-collision domain networks using hierarchical, cascaded repeating hubs to increase port density. 100BASE-T does not permit hierarchical cascading of hubs because the maximum collision domain for UTP is slightly more than 200 meters (see Exhibit 10).
Two techniques can be used to build large, single-collision domain networks (i.e., increase port density). One technique is to use modular hubs, where ports can be added by inserting additional multiport cards into the hub chassis. A second is to use stackable hubs stand-alone repeaters that can be connected via high-bandwidth stacking ports that do not impact the collision domain.
Exhibit 11. 100BASE-T Topologies.
Topology rules for half-duplex 100BASE-T networks are shown in Exhibit 11. Copper links are limited to 100 meters by the U.S. cabling standard EIA/TIA-568-A. A collision domain containing two copper links can contain one class I repeater and two 100-meter copper links; or two class II repeaters, two 100-meter copper links, and a 5-meter copper inter-repeater link. A collision domain containing a class I repeater with two fiber links can support two fiber links of 136 meters, for a collision domain diameter of 272 meters. A collision domain containing a class I repeater can also support one copper link of 100 meters and a single fiber link of 160 meters.
A fiber DTE-DTE half-duplex collision domain (e.g., a switch-to-switch or switch-to-server) can support a 412-meter fiber link. Links of up to 2 kilometers can be supported over fiber by operating in full-duplex mode, which turns off the CSMA/CD portion of the protocol and requires a dedicated link (see Exhibit 12).
Work to extend the Ethernet family to 1000M-bps (gigabit) operation is well underway. The first products using the new technology were demonstrated at Networld + Interop Las Vegas in May 1997, and the first products start shipping during the summer of 1997. Initial products will support operation over 62.5-micron multimode fiber (1000BASE-SX), 50-micron single-mode fiber (1000BASE-LX), or short lengths (to 25 meters) of coaxial cable (1000BASE-CX). The operation of these products is being defined in a supplement to the IEEE 802.3 standard entitled 802.3z
Exhibit 12. Full Duplex.
It is scheduled for completion in 1998. A second supplement, entitled 802.3ab, will define gigabit Ethernet over 100-meter, four-pair Category 5 copper links (1000BASE-T). It is scheduled for completion in late 1998.
Fast Ethernet is a family of 100M-bps signaling systems for use with the standard Ethernet MAC layer. The family consists of four signaling systems (100BASE-TX, 100BASE-T4, 100BASE-T2, and 100BASE-FX) and technologies that support automatic start-up (Auto-Negotiation), shared media operation (Repeaters), full-duplex operation, and flow-control to manage traffic flow.
Fast Ethernet devices work seamlessly with legacy Ethernet systems: they have the same MAC layer, the same frame format, and the same CSMA/CD protocol for shared media operation. Auto-Negotiation ensures that all 100BASE-T devices operating over copper links automatically configure themselves to operate with link partners. This makes 100BASE-T a very economical technology for adding high-bandwidth links to legacy systems.
Higher-speed operation reduces the diameter of 100BASE-T shared media collision domains to approximately 200 meters for copper. Collision domains can be extended through the use of fiber and connected via switches to build large, complex networks. Full-duplex operation improves bandwidth for bidirectional links and increases the maximum length of fiber links to kilometers. Port density within a single collision domain is expanded through the use of modular and stackable hubs.
100M bps is not the endpoint for Ethernet. 1000M-bps (gigabit) devices were demonstrated in the spring of 1997 and began shipping in the summer of 1997. Targets for gigabit Ethernet operation are 700 meters for full-duplex single mode fiber links, 25 meters for short-haul copper coax links, and 100 meters for Category 5 copper links.
Previous | Table of Contents | Next |